WEBVTT Kind: captions Language: en 00:00:00.060 --> 00:00:01.760 align:start position:0% so<00:00:00.390> there's<00:00:00.570> been<00:00:00.750> a<00:00:00.810> lot<00:00:01.170> of<00:00:01.199> excitement 00:00:01.760 --> 00:00:01.770 align:start position:0% so there's been a lot of excitement 00:00:01.770 --> 00:00:05.180 align:start position:0% so there's been a lot of excitement about<00:00:02.070> Prague<00:00:02.490> pal<00:00:02.820> and<00:00:03.350> what<00:00:04.350> is<00:00:04.589> Prague<00:00:04.950> pow 00:00:05.180 --> 00:00:05.190 align:start position:0% about Prague pal and what is Prague pow 00:00:05.190 --> 00:00:07.419 align:start position:0% about Prague pal and what is Prague pow well<00:00:05.520> it's<00:00:05.850> a<00:00:06.089> programmatic<00:00:06.540> proof<00:00:06.990> of<00:00:07.140> work 00:00:07.419 --> 00:00:07.429 align:start position:0% well it's a programmatic proof of work 00:00:07.429 --> 00:00:10.850 align:start position:0% well it's a programmatic proof of work and<00:00:08.780> we're<00:00:09.780> going<00:00:10.019> to<00:00:10.110> be<00:00:10.200> tackling<00:00:10.590> all<00:00:10.740> your 00:00:10.850 --> 00:00:10.860 align:start position:0% and we're going to be tackling all your 00:00:10.860 --> 00:00:13.490 align:start position:0% and we're going to be tackling all your questions<00:00:11.099> and<00:00:11.519> answers<00:00:11.880> today<00:00:12.170> this<00:00:13.170> will<00:00:13.349> be 00:00:13.490 --> 00:00:13.500 align:start position:0% questions and answers today this will be 00:00:13.500 --> 00:00:15.379 align:start position:0% questions and answers today this will be sort<00:00:13.889> of<00:00:13.920> a<00:00:14.009> medium<00:00:14.250> level<00:00:14.700> expertise 00:00:15.379 --> 00:00:15.389 align:start position:0% sort of a medium level expertise 00:00:15.389 --> 00:00:17.720 align:start position:0% sort of a medium level expertise presentation<00:00:16.379> so<00:00:16.619> not<00:00:16.920> too<00:00:17.160> technical<00:00:17.400> but 00:00:17.720 --> 00:00:17.730 align:start position:0% presentation so not too technical but 00:00:17.730 --> 00:00:21.349 align:start position:0% presentation so not too technical but not<00:00:18.000> too<00:00:18.180> generic<00:00:18.359> and<00:00:19.670> if<00:00:20.670> you<00:00:20.880> really<00:00:21.210> want 00:00:21.349 --> 00:00:21.359 align:start position:0% not too generic and if you really want 00:00:21.359 --> 00:00:23.870 align:start position:0% not too generic and if you really want to<00:00:21.420> go<00:00:21.539> in<00:00:21.779> a<00:00:21.869> deeper<00:00:22.199> dive<00:00:22.380> I<00:00:22.800> figure<00:00:23.250> you 00:00:23.870 --> 00:00:23.880 align:start position:0% to go in a deeper dive I figure you 00:00:23.880 --> 00:00:25.400 align:start position:0% to go in a deeper dive I figure you should<00:00:24.090> check<00:00:24.240> out<00:00:24.390> the<00:00:24.510> official<00:00:24.840> github 00:00:25.400 --> 00:00:25.410 align:start position:0% should check out the official github 00:00:25.410 --> 00:00:27.679 align:start position:0% should check out the official github repo<00:00:25.800> with<00:00:26.220> the<00:00:26.490> official<00:00:27.180> code<00:00:27.359> and<00:00:27.630> the 00:00:27.679 --> 00:00:27.689 align:start position:0% repo with the official code and the 00:00:27.689 --> 00:00:29.839 align:start position:0% repo with the official code and the white<00:00:27.869> paper<00:00:28.109> as<00:00:28.470> well<00:00:29.220> as<00:00:29.340> our<00:00:29.490> medium 00:00:29.839 --> 00:00:29.849 align:start position:0% white paper as well as our medium 00:00:29.849 --> 00:00:34.940 align:start position:0% white paper as well as our medium accounts<00:00:31.520> but<00:00:32.520> first<00:00:32.850> a<00:00:33.329> little<00:00:33.960> bit<00:00:34.170> about 00:00:34.940 --> 00:00:34.950 align:start position:0% accounts but first a little bit about 00:00:34.950 --> 00:00:38.660 align:start position:0% accounts but first a little bit about the<00:00:35.610> team<00:00:36.079> so<00:00:37.079> too<00:00:37.680> many<00:00:37.890> people<00:00:38.309> put<00:00:38.489> an 00:00:38.660 --> 00:00:38.670 align:start position:0% the team so too many people put an 00:00:38.670 --> 00:00:41.360 align:start position:0% the team so too many people put an emphasis<00:00:38.820> on<00:00:39.300> the<00:00:39.450> people<00:00:39.719> behind<00:00:40.370> algorithms 00:00:41.360 --> 00:00:41.370 align:start position:0% emphasis on the people behind algorithms 00:00:41.370 --> 00:00:43.610 align:start position:0% emphasis on the people behind algorithms or<00:00:41.550> work<00:00:41.790> which<00:00:42.059> is<00:00:42.239> highly<00:00:42.899> amusing<00:00:43.230> to<00:00:43.500> me 00:00:43.610 --> 00:00:43.620 align:start position:0% or work which is highly amusing to me 00:00:43.620 --> 00:00:45.760 align:start position:0% or work which is highly amusing to me especially<00:00:43.739> considering<00:00:44.700> that<00:00:44.820> you<00:00:45.510> know 00:00:45.760 --> 00:00:45.770 align:start position:0% especially considering that you know 00:00:45.770 --> 00:00:49.910 align:start position:0% especially considering that you know cryptocurrency<00:00:46.770> is<00:00:47.129> all<00:00:47.280> about<00:00:47.550> no<00:00:48.480> trust<00:00:48.920> but 00:00:49.910 --> 00:00:49.920 align:start position:0% cryptocurrency is all about no trust but 00:00:49.920 --> 00:00:52.639 align:start position:0% cryptocurrency is all about no trust but hey<00:00:50.129> I<00:00:50.399> learned<00:00:51.239> from<00:00:51.870> the<00:00:51.989> first<00:00:52.170> time<00:00:52.410> when 00:00:52.639 --> 00:00:52.649 align:start position:0% hey I learned from the first time when 00:00:52.649 --> 00:00:54.229 align:start position:0% hey I learned from the first time when we<00:00:52.680> were<00:00:52.829> released<00:00:53.100> Prague<00:00:53.399> pal<00:00:53.610> that<00:00:53.879> you've 00:00:54.229 --> 00:00:54.239 align:start position:0% we were released Prague pal that you've 00:00:54.239 --> 00:00:56.299 align:start position:0% we were released Prague pal that you've actually<00:00:54.570> have<00:00:54.719> to<00:00:54.809> do<00:00:54.899> these<00:00:55.079> disclaimers<00:00:55.530> so 00:00:56.299 --> 00:00:56.309 align:start position:0% actually have to do these disclaimers so 00:00:56.309 --> 00:00:59.900 align:start position:0% actually have to do these disclaimers so I<00:00:56.789> figure<00:00:57.800> we<00:00:58.800> have<00:00:58.949> to<00:00:59.070> do<00:00:59.190> a<00:00:59.219> disclosure<00:00:59.640> on 00:00:59.900 --> 00:00:59.910 align:start position:0% I figure we have to do a disclosure on 00:00:59.910 --> 00:01:03.500 align:start position:0% I figure we have to do a disclosure on who<00:01:00.090> we<00:01:00.239> are<00:01:00.530> so<00:01:01.530> it<00:01:01.800> Pro<00:01:02.579> POW<00:01:02.820> is<00:01:03.090> made<00:01:03.270> by 00:01:03.500 --> 00:01:03.510 align:start position:0% who we are so it Pro POW is made by 00:01:03.510 --> 00:01:07.490 align:start position:0% who we are so it Pro POW is made by three<00:01:04.019> people<00:01:04.320> if<00:01:04.530> def<00:01:04.949> else<00:01:05.189> I<00:01:05.630> am<00:01:06.630> miss<00:01:06.869> if<00:01:07.110> I 00:01:07.490 --> 00:01:07.500 align:start position:0% three people if def else I am miss if I 00:01:07.500 --> 00:01:10.039 align:start position:0% three people if def else I am miss if I am<00:01:08.189> currently<00:01:08.430> chief<00:01:08.850> technology<00:01:09.450> officer<00:01:09.900> of 00:01:10.039 --> 00:01:10.049 align:start position:0% am currently chief technology officer of 00:01:10.049 --> 00:01:12.289 align:start position:0% am currently chief technology officer of course<00:01:10.260> scientific<00:01:10.950> course<00:01:11.580> scientific<00:01:12.180> is 00:01:12.289 --> 00:01:12.299 align:start position:0% course scientific course scientific is 00:01:12.299 --> 00:01:14.450 align:start position:0% course scientific course scientific is an<00:01:12.390> artificial<00:01:13.159> intelligence<00:01:14.159> and 00:01:14.450 --> 00:01:14.460 align:start position:0% an artificial intelligence and 00:01:14.460 --> 00:01:16.070 align:start position:0% an artificial intelligence and blockchain<00:01:14.850> infrastructure<00:01:15.600> and<00:01:15.720> hosting 00:01:16.070 --> 00:01:16.080 align:start position:0% blockchain infrastructure and hosting 00:01:16.080 --> 00:01:18.440 align:start position:0% blockchain infrastructure and hosting company<00:01:16.409> I'm<00:01:17.310> also<00:01:17.580> the<00:01:17.850> founder<00:01:18.030> of<00:01:18.330> the 00:01:18.440 --> 00:01:18.450 align:start position:0% company I'm also the founder of the 00:01:18.450 --> 00:01:20.270 align:start position:0% company I'm also the founder of the minority<00:01:18.960> group<00:01:19.229> who<00:01:19.500> is<00:01:19.650> responsible<00:01:20.250> for 00:01:20.270 --> 00:01:20.280 align:start position:0% minority group who is responsible for 00:01:20.280 --> 00:01:23.300 align:start position:0% minority group who is responsible for our<00:01:20.549> fantastic<00:01:21.150> graphics<00:01:21.360> today<00:01:21.869> and<00:01:22.310> I'm 00:01:23.300 --> 00:01:23.310 align:start position:0% our fantastic graphics today and I'm 00:01:23.310 --> 00:01:25.310 align:start position:0% our fantastic graphics today and I'm also<00:01:23.520> extraneous<00:01:24.090> mining<00:01:24.659> the<00:01:24.990> world's 00:01:25.310 --> 00:01:25.320 align:start position:0% also extraneous mining the world's 00:01:25.320 --> 00:01:28.850 align:start position:0% also extraneous mining the world's leading<00:01:25.560> hashing<00:01:26.220> power<00:01:26.250> provider<00:01:26.850> and<00:01:27.860> then 00:01:28.850 --> 00:01:28.860 align:start position:0% leading hashing power provider and then 00:01:28.860 --> 00:01:31.969 align:start position:0% leading hashing power provider and then we<00:01:29.250> also<00:01:29.430> have<00:01:29.759> mr.<00:01:30.090> def<00:01:30.450> mr.<00:01:31.229> def<00:01:31.409> is<00:01:31.650> an 00:01:31.969 --> 00:01:31.979 align:start position:0% we also have mr. def mr. def is an 00:01:31.979 --> 00:01:34.100 align:start position:0% we also have mr. def mr. def is an experienced<00:01:32.610> systems<00:01:33.060> engineer<00:01:33.540> he's<00:01:34.079> a 00:01:34.100 --> 00:01:34.110 align:start position:0% experienced systems engineer he's a 00:01:34.110 --> 00:01:36.289 align:start position:0% experienced systems engineer he's a blockchain<00:01:34.439> enthusiast<00:01:35.280> and<00:01:35.790> I<00:01:35.909> like<00:01:36.060> to<00:01:36.180> call 00:01:36.289 --> 00:01:36.299 align:start position:0% blockchain enthusiast and I like to call 00:01:36.299 --> 00:01:38.120 align:start position:0% blockchain enthusiast and I like to call him<00:01:36.390> a<00:01:36.540> professional<00:01:37.049> cat<00:01:37.229> herder<00:01:37.619> because 00:01:38.120 --> 00:01:38.130 align:start position:0% him a professional cat herder because 00:01:38.130 --> 00:01:41.149 align:start position:0% him a professional cat herder because that's<00:01:38.340> pretty<00:01:38.490> much<00:01:38.610> what<00:01:38.780> 24<00:01:39.780> 24<00:01:40.770> hours<00:01:41.009> of 00:01:41.149 --> 00:01:41.159 align:start position:0% that's pretty much what 24 24 hours of 00:01:41.159 --> 00:01:44.359 align:start position:0% that's pretty much what 24 24 hours of his<00:01:41.280> job<00:01:41.579> consists<00:01:42.090> of<00:01:42.619> and<00:01:43.619> then<00:01:43.740> we<00:01:43.829> have<00:01:43.979> mr. 00:01:44.359 --> 00:01:44.369 align:start position:0% his job consists of and then we have mr. 00:01:44.369 --> 00:01:45.620 align:start position:0% his job consists of and then we have mr. mr.<00:01:45.119> elfs 00:01:45.620 --> 00:01:45.630 align:start position:0% mr. elfs 00:01:45.630 --> 00:01:48.139 align:start position:0% mr. elfs so<00:01:46.140> mr.<00:01:46.470> else<00:01:46.649> is<00:01:46.890> quite<00:01:47.340> an<00:01:47.490> experienced 00:01:48.139 --> 00:01:48.149 align:start position:0% so mr. else is quite an experienced 00:01:48.149 --> 00:01:50.719 align:start position:0% so mr. else is quite an experienced semiconductor<00:01:48.899> engineer<00:01:49.350> he<00:01:49.950> has<00:01:50.070> over<00:01:50.369> 10 00:01:50.719 --> 00:01:50.729 align:start position:0% semiconductor engineer he has over 10 00:01:50.729 --> 00:01:53.359 align:start position:0% semiconductor engineer he has over 10 years<00:01:51.000> of<00:01:51.149> experience<00:01:51.270> working<00:01:52.159> in<00:01:53.159> this 00:01:53.359 --> 00:01:53.369 align:start position:0% years of experience working in this 00:01:53.369 --> 00:01:58.639 align:start position:0% years of experience working in this space<00:01:53.549> and<00:01:55.880> he<00:01:56.880> is<00:01:57.180> also<00:01:57.479> an<00:01:57.689> experienced<00:01:58.229> GPU 00:01:58.639 --> 00:01:58.649 align:start position:0% space and he is also an experienced GPU 00:01:58.649 --> 00:02:01.940 align:start position:0% space and he is also an experienced GPU architect<00:01:59.250> and<00:01:59.520> he<00:01:59.790> loves<00:02:00.030> Prague<00:02:00.540> pal<00:02:00.950> but 00:02:01.940 --> 00:02:01.950 align:start position:0% architect and he loves Prague pal but 00:02:01.950 --> 00:02:03.859 align:start position:0% architect and he loves Prague pal but not<00:02:02.340> the<00:02:02.430> pro<00:02:02.610> pal<00:02:02.850> we're<00:02:03.060> all<00:02:03.210> familiar<00:02:03.390> with 00:02:03.859 --> 00:02:03.869 align:start position:0% not the pro pal we're all familiar with 00:02:03.869 --> 00:02:09.859 align:start position:0% not the pro pal we're all familiar with he<00:02:04.799> loves<00:02:05.009> this<00:02:05.159> program<00:02:06.890> and<00:02:08.330> together<00:02:09.330> we 00:02:09.859 --> 00:02:09.869 align:start position:0% he loves this program and together we 00:02:09.869 --> 00:02:11.460 align:start position:0% he loves this program and together we are<00:02:09.899> if<00:02:10.140> def<00:02:10.379> else 00:02:11.460 --> 00:02:11.470 align:start position:0% are if def else 00:02:11.470 --> 00:02:16.980 align:start position:0% are if def else so<00:02:12.340> what<00:02:12.520> is<00:02:12.700> prog<00:02:13.060> pal<00:02:14.970> well<00:02:15.970> as<00:02:16.510> I<00:02:16.750> said 00:02:16.980 --> 00:02:16.990 align:start position:0% so what is prog pal well as I said 00:02:16.990 --> 00:02:18.900 align:start position:0% so what is prog pal well as I said before<00:02:17.140> it's<00:02:17.590> a<00:02:17.740> programmable<00:02:18.670> a 00:02:18.900 --> 00:02:18.910 align:start position:0% before it's a programmable a 00:02:18.910 --> 00:02:22.140 align:start position:0% before it's a programmable a programmatic<00:02:19.300> proof-of-work<00:02:20.260> or<00:02:20.920> as<00:02:21.670> we<00:02:21.970> like 00:02:22.140 --> 00:02:22.150 align:start position:0% programmatic proof-of-work or as we like 00:02:22.150 --> 00:02:24.300 align:start position:0% programmatic proof-of-work or as we like to<00:02:22.180> say<00:02:22.300> a<00:02:22.420> GP<00:02:22.870> you<00:02:22.900> tuned<00:02:23.290> extension<00:02:23.980> of<00:02:24.130> it 00:02:24.300 --> 00:02:24.310 align:start position:0% to say a GP you tuned extension of it 00:02:24.310 --> 00:02:28.440 align:start position:0% to say a GP you tuned extension of it hash<00:02:25.260> it's<00:02:26.260> also<00:02:26.470> sometimes<00:02:26.890> called<00:02:27.450> portal 00:02:28.440 --> 00:02:28.450 align:start position:0% hash it's also sometimes called portal 00:02:28.450 --> 00:02:32.550 align:start position:0% hash it's also sometimes called portal or<00:02:28.720> pongy<00:02:29.260> well<00:02:29.410> after<00:02:30.280> this<00:02:30.820> little<00:02:31.150> guy<00:02:31.560> our 00:02:32.550 --> 00:02:32.560 align:start position:0% or pongy well after this little guy our 00:02:32.560 --> 00:02:35.390 align:start position:0% or pongy well after this little guy our little<00:02:32.920> hero<00:02:33.190> from<00:02:33.580> Star<00:02:33.850> Wars<00:02:33.880> lost<00:02:34.420> Jedi<00:02:34.810> and 00:02:35.390 --> 00:02:35.400 align:start position:0% little hero from Star Wars lost Jedi and 00:02:35.400 --> 00:02:37.890 align:start position:0% little hero from Star Wars lost Jedi and it<00:02:36.400> also<00:02:36.550> takes<00:02:36.880> its<00:02:37.120> roots<00:02:37.360> from<00:02:37.510> the<00:02:37.660> word 00:02:37.890 --> 00:02:37.900 align:start position:0% it also takes its roots from the word 00:02:37.900 --> 00:02:40.770 align:start position:0% it also takes its roots from the word programmatic<00:02:38.620> now<00:02:39.580> I<00:02:40.000> quite<00:02:40.570> like 00:02:40.770 --> 00:02:40.780 align:start position:0% programmatic now I quite like 00:02:40.780 --> 00:02:42.120 align:start position:0% programmatic now I quite like programmatic<00:02:41.410> because<00:02:41.560> it's<00:02:41.680> a<00:02:41.740> great<00:02:41.920> way<00:02:42.100> to 00:02:42.120 --> 00:02:42.130 align:start position:0% programmatic because it's a great way to 00:02:42.130 --> 00:02:44.850 align:start position:0% programmatic because it's a great way to explain<00:02:42.850> algorithms<00:02:43.720> algorithms<00:02:44.470> are 00:02:44.850 --> 00:02:44.860 align:start position:0% explain algorithms algorithms are 00:02:44.860 --> 00:02:47.550 align:start position:0% explain algorithms algorithms are actually<00:02:45.160> very<00:02:45.490> musical<00:02:46.120> I'm<00:02:46.690> you<00:02:47.080> know<00:02:47.200> every 00:02:47.550 --> 00:02:47.560 align:start position:0% actually very musical I'm you know every 00:02:47.560 --> 00:02:51.180 align:start position:0% actually very musical I'm you know every algorithm<00:02:48.220> has<00:02:48.880> a<00:02:49.240> beat<00:02:49.540> and<00:02:50.190> optimization 00:02:51.180 --> 00:02:51.190 align:start position:0% algorithm has a beat and optimization 00:02:51.190 --> 00:02:52.830 align:start position:0% algorithm has a beat and optimization when<00:02:51.550> we<00:02:51.640> talk<00:02:51.820> about<00:02:51.850> optimization<00:02:52.210> of 00:02:52.830 --> 00:02:52.840 align:start position:0% when we talk about optimization of 00:02:52.840 --> 00:02:54.660 align:start position:0% when we talk about optimization of software<00:02:53.320> or<00:02:53.440> hardware<00:02:53.890> it's<00:02:54.370> all<00:02:54.490> about 00:02:54.660 --> 00:02:54.670 align:start position:0% software or hardware it's all about 00:02:54.670 --> 00:02:56.730 align:start position:0% software or hardware it's all about matching<00:02:55.150> the<00:02:55.270> beat<00:02:55.480> of<00:02:55.630> an<00:02:55.720> algorithm<00:02:56.020> to<00:02:56.620> the 00:02:56.730 --> 00:02:56.740 align:start position:0% matching the beat of an algorithm to the 00:02:56.740 --> 00:02:59.670 align:start position:0% matching the beat of an algorithm to the beat<00:02:56.920> of<00:02:56.950> a<00:02:57.130> hardware<00:02:57.450> so<00:02:58.450> at<00:02:58.690> hash<00:02:58.960> I<00:02:59.320> like<00:02:59.560> to 00:02:59.670 --> 00:02:59.680 align:start position:0% beat of a hardware so at hash I like to 00:02:59.680 --> 00:03:01.980 align:start position:0% beat of a hardware so at hash I like to say<00:02:59.800> is<00:02:59.920> a<00:02:59.950> little<00:03:00.070> like<00:03:00.310> trap<00:03:00.580> core<00:03:00.880> Manero<00:03:01.780> is 00:03:01.980 --> 00:03:01.990 align:start position:0% say is a little like trap core Manero is 00:03:01.990 --> 00:03:04.110 align:start position:0% say is a little like trap core Manero is like<00:03:02.140> dubstep<00:03:02.410> z<00:03:03.280> cache<00:03:03.520> is<00:03:03.700> a<00:03:03.730> little<00:03:04.000> like 00:03:04.110 --> 00:03:04.120 align:start position:0% like dubstep z cache is a little like 00:03:04.120 --> 00:03:07.200 align:start position:0% like dubstep z cache is a little like rock<00:03:04.420> and<00:03:04.780> these<00:03:05.620> beats<00:03:05.950> like<00:03:06.670> I<00:03:06.820> said<00:03:07.030> they 00:03:07.200 --> 00:03:07.210 align:start position:0% rock and these beats like I said they 00:03:07.210 --> 00:03:09.210 align:start position:0% rock and these beats like I said they need<00:03:07.240> to<00:03:07.510> synchronize<00:03:08.320> with<00:03:08.350> the<00:03:08.860> hardware 00:03:09.210 --> 00:03:09.220 align:start position:0% need to synchronize with the hardware 00:03:09.220 --> 00:03:14.610 align:start position:0% need to synchronize with the hardware itself<00:03:10.560> so<00:03:11.560> let's<00:03:11.680> do<00:03:11.950> a<00:03:12.160> big<00:03:12.550> deep<00:03:12.760> dive<00:03:13.620> right 00:03:14.610 --> 00:03:14.620 align:start position:0% itself so let's do a big deep dive right 00:03:14.620 --> 00:03:17.970 align:start position:0% itself so let's do a big deep dive right now<00:03:14.830> why<00:03:15.190> was<00:03:15.400> prog<00:03:15.670> pal<00:03:15.910> created<00:03:16.600> well<00:03:17.080> today 00:03:17.970 --> 00:03:17.980 align:start position:0% now why was prog pal created well today 00:03:17.980 --> 00:03:19.590 align:start position:0% now why was prog pal created well today there's<00:03:18.280> kind<00:03:18.730> of<00:03:18.850> a<00:03:18.970> problem<00:03:19.209> with 00:03:19.590 --> 00:03:19.600 align:start position:0% there's kind of a problem with 00:03:19.600 --> 00:03:23.390 align:start position:0% there's kind of a problem with algorithms<00:03:20.230> you<00:03:21.160> see<00:03:21.420> traditionally 00:03:23.390 --> 00:03:23.400 align:start position:0% algorithms you see traditionally 00:03:23.400 --> 00:03:25.860 align:start position:0% algorithms you see traditionally algorithms<00:03:24.400> are<00:03:24.610> designed<00:03:24.850> by<00:03:25.480> software 00:03:25.860 --> 00:03:25.870 align:start position:0% algorithms are designed by software 00:03:25.870 --> 00:03:29.340 align:start position:0% algorithms are designed by software engineers<00:03:26.790> now<00:03:27.790> you<00:03:28.360> might<00:03:28.540> be<00:03:28.570> asking<00:03:28.810> what's 00:03:29.340 --> 00:03:29.350 align:start position:0% engineers now you might be asking what's 00:03:29.350 --> 00:03:31.979 align:start position:0% engineers now you might be asking what's wrong<00:03:29.620> with<00:03:29.770> that<00:03:29.830> well<00:03:30.510> would<00:03:31.510> you<00:03:31.630> want<00:03:31.870> your 00:03:31.979 --> 00:03:31.989 align:start position:0% wrong with that well would you want your 00:03:31.989 --> 00:03:34.160 align:start position:0% wrong with that well would you want your hardware<00:03:32.410> designed<00:03:32.830> by<00:03:33.070> a<00:03:33.100> software<00:03:33.610> engineer 00:03:34.160 --> 00:03:34.170 align:start position:0% hardware designed by a software engineer 00:03:34.170 --> 00:03:37.620 align:start position:0% hardware designed by a software engineer the<00:03:35.170> answer<00:03:35.500> is<00:03:35.650> no<00:03:35.980> because<00:03:36.420> that's<00:03:37.420> actually 00:03:37.620 --> 00:03:37.630 align:start position:0% the answer is no because that's actually 00:03:37.630 --> 00:03:40.160 align:start position:0% the answer is no because that's actually how<00:03:37.989> you<00:03:38.050> end<00:03:38.320> up<00:03:38.500> with<00:03:38.770> things<00:03:39.160> like<00:03:39.459> these 00:03:40.160 --> 00:03:40.170 align:start position:0% how you end up with things like these 00:03:40.170 --> 00:03:45.870 align:start position:0% how you end up with things like these the<00:03:41.170> windows<00:03:41.500> phone<00:03:43.230> so<00:03:44.230> we<00:03:44.980> we<00:03:45.340> always<00:03:45.670> talk 00:03:45.870 --> 00:03:45.880 align:start position:0% the windows phone so we we always talk 00:03:45.880 --> 00:03:48.000 align:start position:0% the windows phone so we we always talk about<00:03:46.150> how<00:03:46.390> you<00:03:46.450> need<00:03:46.959> to<00:03:47.019> have<00:03:47.410> people<00:03:47.890> who 00:03:48.000 --> 00:03:48.010 align:start position:0% about how you need to have people who 00:03:48.010 --> 00:03:49.830 align:start position:0% about how you need to have people who are<00:03:48.130> experts<00:03:48.700> left<00:03:49.060> to<00:03:49.180> do<00:03:49.330> the<00:03:49.510> things<00:03:49.720> that 00:03:49.830 --> 00:03:49.840 align:start position:0% are experts left to do the things that 00:03:49.840 --> 00:03:51.630 align:start position:0% are experts left to do the things that they're<00:03:50.050> an<00:03:50.140> expert<00:03:50.380> in<00:03:50.650> how<00:03:51.430> do<00:03:51.489> our 00:03:51.630 --> 00:03:51.640 align:start position:0% they're an expert in how do our 00:03:51.640 --> 00:03:54.270 align:start position:0% they're an expert in how do our engineers<00:03:52.150> so<00:03:53.050> it<00:03:53.110> should<00:03:53.470> tune<00:03:53.739> for<00:03:53.950> hardware 00:03:54.270 --> 00:03:54.280 align:start position:0% engineers so it should tune for hardware 00:03:54.280 --> 00:03:56.990 align:start position:0% engineers so it should tune for hardware soft<00:03:54.910> engine<00:03:55.209> software<00:03:55.540> engineers<00:03:56.230> should 00:03:56.990 --> 00:03:57.000 align:start position:0% soft engine software engineers should 00:03:57.000 --> 00:04:00.360 align:start position:0% soft engine software engineers should tune<00:03:58.000> for<00:03:58.209> application<00:03:58.780> layers<00:03:59.100> that's<00:04:00.100> the 00:04:00.360 --> 00:04:00.370 align:start position:0% tune for application layers that's the 00:04:00.370 --> 00:04:02.580 align:start position:0% tune for application layers that's the crux<00:04:00.640> of<00:04:00.820> the<00:04:00.880> problem<00:04:00.970> today<00:04:01.480> and<00:04:01.840> how<00:04:02.260> we<00:04:02.320> get 00:04:02.580 --> 00:04:02.590 align:start position:0% crux of the problem today and how we get 00:04:02.590 --> 00:04:04.979 align:start position:0% crux of the problem today and how we get fixed<00:04:03.010> function<00:04:03.190> hardware<00:04:03.760> or<00:04:03.940> in<00:04:04.300> this<00:04:04.420> case 00:04:04.979 --> 00:04:04.989 align:start position:0% fixed function hardware or in this case 00:04:04.989 --> 00:04:08.310 align:start position:0% fixed function hardware or in this case Asics<00:04:05.680> and<00:04:06.930> proof-of-work<00:04:07.930> has 00:04:08.310 --> 00:04:08.320 align:start position:0% Asics and proof-of-work has 00:04:08.320 --> 00:04:11.370 align:start position:0% Asics and proof-of-work has traditionally<00:04:08.860> taken<00:04:09.160> an<00:04:09.400> algorithm<00:04:10.380> that's 00:04:11.370 --> 00:04:11.380 align:start position:0% traditionally taken an algorithm that's 00:04:11.380 --> 00:04:13.440 align:start position:0% traditionally taken an algorithm that's fixed<00:04:11.950> and<00:04:12.100> tried<00:04:12.430> to<00:04:12.610> shoehorn<00:04:12.880> in<00:04:13.330> the 00:04:13.440 --> 00:04:13.450 align:start position:0% fixed and tried to shoehorn in the 00:04:13.450 --> 00:04:14.820 align:start position:0% fixed and tried to shoehorn in the hardware<00:04:13.840> to<00:04:14.050> make<00:04:14.200> it<00:04:14.320> efficient<00:04:14.739> at 00:04:14.820 --> 00:04:14.830 align:start position:0% hardware to make it efficient at 00:04:14.830 --> 00:04:17.670 align:start position:0% hardware to make it efficient at executing<00:04:15.489> it<00:04:15.610> but<00:04:16.359> that<00:04:16.480> leaves<00:04:16.690> many<00:04:17.080> unused 00:04:17.670 --> 00:04:17.680 align:start position:0% executing it but that leaves many unused 00:04:17.680 --> 00:04:19.529 align:start position:0% executing it but that leaves many unused parts<00:04:18.220> of<00:04:18.430> the<00:04:18.549> hardware<00:04:18.940> that<00:04:19.090> you<00:04:19.209> can<00:04:19.359> just 00:04:19.529 --> 00:04:19.539 align:start position:0% parts of the hardware that you can just 00:04:19.539 --> 00:04:23.219 align:start position:0% parts of the hardware that you can just shave<00:04:19.780> off<00:04:20.220> or<00:04:21.220> you<00:04:21.370> can<00:04:21.549> get<00:04:21.700> rid<00:04:21.880> of<00:04:22.229> instead 00:04:23.219 --> 00:04:23.229 align:start position:0% shave off or you can get rid of instead 00:04:23.229 --> 00:04:26.510 align:start position:0% shave off or you can get rid of instead Pro<00:04:23.770> pal<00:04:24.010> flips<00:04:24.550> this<00:04:24.700> paradigm 00:04:26.510 --> 00:04:26.520 align:start position:0% Pro pal flips this paradigm 00:04:26.520 --> 00:04:29.790 align:start position:0% Pro pal flips this paradigm we<00:04:27.520> take<00:04:27.789> the<00:04:28.090> hardware<00:04:28.800> already-existing 00:04:29.790 --> 00:04:29.800 align:start position:0% we take the hardware already-existing 00:04:29.800 --> 00:04:32.040 align:start position:0% we take the hardware already-existing and<00:04:30.340> we<00:04:30.430> modify<00:04:30.849> an<00:04:30.970> algorithm<00:04:31.509> to<00:04:31.539> match<00:04:31.810> it 00:04:32.040 --> 00:04:32.050 align:start position:0% and we modify an algorithm to match it 00:04:32.050 --> 00:04:34.409 align:start position:0% and we modify an algorithm to match it because<00:04:32.919> an<00:04:33.039> efficient<00:04:33.490> algorithm<00:04:34.060> for<00:04:34.090> proof 00:04:34.409 --> 00:04:34.419 align:start position:0% because an efficient algorithm for proof 00:04:34.419 --> 00:04:36.900 align:start position:0% because an efficient algorithm for proof of<00:04:34.539> work<00:04:34.720> means<00:04:35.410> hardware<00:04:35.919> needs<00:04:36.250> to<00:04:36.460> match<00:04:36.639> s 00:04:36.900 --> 00:04:36.910 align:start position:0% of work means hardware needs to match s 00:04:36.910 --> 00:04:38.939 align:start position:0% of work means hardware needs to match s match<00:04:37.389> access<00:04:37.960> patterns<00:04:38.380> and<00:04:38.560> available 00:04:38.939 --> 00:04:38.949 align:start position:0% match access patterns and available 00:04:38.949 --> 00:04:42.930 align:start position:0% match access patterns and available space<00:04:39.220> of<00:04:39.400> that<00:04:39.520> hardware<00:04:40.620> now<00:04:41.620> I<00:04:41.979> know<00:04:42.580> what 00:04:42.930 --> 00:04:42.940 align:start position:0% space of that hardware now I know what 00:04:42.940 --> 00:04:44.909 align:start position:0% space of that hardware now I know what everyone<00:04:43.539> right<00:04:43.780> now<00:04:43.960> is<00:04:44.229> waiting<00:04:44.470> for<00:04:44.710> me<00:04:44.800> to 00:04:44.909 --> 00:04:44.919 align:start position:0% everyone right now is waiting for me to 00:04:44.919 --> 00:04:47.159 align:start position:0% everyone right now is waiting for me to say<00:04:45.130> given<00:04:45.639> how<00:04:45.759> vocal<00:04:46.330> I<00:04:46.419> am<00:04:46.569> about<00:04:46.780> ASIC 00:04:47.159 --> 00:04:47.169 align:start position:0% say given how vocal I am about ASIC 00:04:47.169 --> 00:04:51.620 align:start position:0% say given how vocal I am about ASIC resistance<00:04:48.090> but<00:04:49.090> I<00:04:49.720> hate<00:04:50.139> to<00:04:50.289> say<00:04:50.470> it 00:04:51.620 --> 00:04:51.630 align:start position:0% resistance but I hate to say it 00:04:51.630 --> 00:04:54.540 align:start position:0% resistance but I hate to say it ASIC<00:04:52.630> resistance<00:04:53.139> is<00:04:53.259> a<00:04:53.289> fallacy<00:04:53.680> there's<00:04:54.370> no 00:04:54.540 --> 00:04:54.550 align:start position:0% ASIC resistance is a fallacy there's no 00:04:54.550 --> 00:04:57.930 align:start position:0% ASIC resistance is a fallacy there's no such<00:04:54.819> thing<00:04:55.770> C<00:04:56.770> prog<00:04:57.039> how<00:04:57.220> is<00:04:57.400> mistakenly 00:04:57.930 --> 00:04:57.940 align:start position:0% such thing C prog how is mistakenly 00:04:57.940 --> 00:05:00.270 align:start position:0% such thing C prog how is mistakenly trumpeted<00:04:58.479> as<00:04:58.599> an<00:04:58.840> ASIC<00:04:59.199> resistant<00:04:59.710> algorithm 00:05:00.270 --> 00:05:00.280 align:start position:0% trumpeted as an ASIC resistant algorithm 00:05:00.280 --> 00:05:03.140 align:start position:0% trumpeted as an ASIC resistant algorithm but<00:05:00.849> quite<00:05:01.569> frankly<00:05:01.599> that's 00:05:03.140 --> 00:05:03.150 align:start position:0% but quite frankly that's 00:05:03.150 --> 00:05:06.480 align:start position:0% but quite frankly that's proof<00:05:04.150> of<00:05:04.300> work<00:05:04.539> requires<00:05:05.199> some<00:05:05.949> form<00:05:06.250> of<00:05:06.400> an 00:05:06.480 --> 00:05:06.490 align:start position:0% proof of work requires some form of an 00:05:06.490 --> 00:05:10.439 align:start position:0% proof of work requires some form of an ASIC<00:05:06.909> to<00:05:07.030> do<00:05:07.240> work<00:05:07.770> be<00:05:08.770> it<00:05:08.860> be<00:05:09.400> it<00:05:09.490> a<00:05:09.580> CPU<00:05:10.060> a<00:05:10.300> GPU 00:05:10.439 --> 00:05:10.449 align:start position:0% ASIC to do work be it be it a CPU a GPU 00:05:10.449 --> 00:05:14.360 align:start position:0% ASIC to do work be it be it a CPU a GPU and<00:05:11.169> FPGA<00:05:11.710> or<00:05:11.830> mining<00:05:12.520> ASIC<00:05:12.880> and<00:05:13.120> guess<00:05:13.900> what 00:05:14.360 --> 00:05:14.370 align:start position:0% and FPGA or mining ASIC and guess what 00:05:14.370 --> 00:05:20.100 align:start position:0% and FPGA or mining ASIC and guess what oops<00:05:17.039> they're<00:05:18.039> all<00:05:18.190> Asics<00:05:18.720> they're<00:05:19.720> all<00:05:19.840> built 00:05:20.100 --> 00:05:20.110 align:start position:0% oops they're all Asics they're all built 00:05:20.110 --> 00:05:23.070 align:start position:0% oops they're all Asics they're all built using<00:05:20.560> the<00:05:20.860> exact<00:05:21.220> same<00:05:21.250> fabs<00:05:21.729> the<00:05:22.569> exact<00:05:22.900> same 00:05:23.070 --> 00:05:23.080 align:start position:0% using the exact same fabs the exact same 00:05:23.080 --> 00:05:25.230 align:start position:0% using the exact same fabs the exact same technologies<00:05:23.770> the<00:05:24.190> exact<00:05:24.490> same<00:05:24.699> materials 00:05:25.230 --> 00:05:25.240 align:start position:0% technologies the exact same materials 00:05:25.240 --> 00:05:28.170 align:start position:0% technologies the exact same materials the<00:05:26.169> difference<00:05:26.560> however<00:05:26.770> in<00:05:27.280> the<00:05:27.699> in<00:05:27.940> this 00:05:28.170 --> 00:05:28.180 align:start position:0% the difference however in the in this 00:05:28.180 --> 00:05:30.629 align:start position:0% the difference however in the in this hardware<00:05:28.630> is<00:05:28.780> the<00:05:29.169> class<00:05:29.620> of<00:05:29.860> algorithms 00:05:30.629 --> 00:05:30.639 align:start position:0% hardware is the class of algorithms 00:05:30.639 --> 00:05:32.610 align:start position:0% hardware is the class of algorithms they're<00:05:30.849> designed<00:05:31.120> to<00:05:31.479> execute<00:05:32.320> efficiently 00:05:32.610 --> 00:05:32.620 align:start position:0% they're designed to execute efficiently 00:05:32.620 --> 00:05:35.070 align:start position:0% they're designed to execute efficiently and<00:05:33.099> that<00:05:33.280> they're<00:05:33.430> optimized<00:05:33.909> for<00:05:34.180> so 00:05:35.070 --> 00:05:35.080 align:start position:0% and that they're optimized for so 00:05:35.080 --> 00:05:36.930 align:start position:0% and that they're optimized for so generally<00:05:35.500> when<00:05:35.620> people<00:05:35.680> say<00:05:36.190> when<00:05:36.639> they<00:05:36.759> say 00:05:36.930 --> 00:05:36.940 align:start position:0% generally when people say when they say 00:05:36.940 --> 00:05:38.640 align:start position:0% generally when people say when they say ASIC<00:05:37.300> resistance<00:05:37.810> what<00:05:38.050> their<00:05:38.229> meaning<00:05:38.530> is 00:05:38.640 --> 00:05:38.650 align:start position:0% ASIC resistance what their meaning is 00:05:38.650 --> 00:05:41.129 align:start position:0% ASIC resistance what their meaning is centralization<00:05:39.490> resistance<00:05:40.060> and<00:05:40.599> we<00:05:40.990> can 00:05:41.129 --> 00:05:41.139 align:start position:0% centralization resistance and we can 00:05:41.139 --> 00:05:43.649 align:start position:0% centralization resistance and we can dive<00:05:41.349> into<00:05:41.620> why<00:05:41.949> that's<00:05:42.190> important<00:05:42.820> in<00:05:43.509> the 00:05:43.649 --> 00:05:43.659 align:start position:0% dive into why that's important in the 00:05:43.659 --> 00:05:47.249 align:start position:0% dive into why that's important in the breakout<00:05:44.020> session<00:05:44.500> but<00:05:46.020> there's<00:05:47.020> my 00:05:47.249 --> 00:05:47.259 align:start position:0% breakout session but there's my 00:05:47.259 --> 00:05:48.870 align:start position:0% breakout session but there's my statement<00:05:47.740> on<00:05:47.770> ASIC<00:05:48.159> resistance 00:05:48.870 --> 00:05:48.880 align:start position:0% statement on ASIC resistance 00:05:48.880 --> 00:05:53.399 align:start position:0% statement on ASIC resistance it<00:05:49.090> doesn't<00:05:49.539> actually<00:05:49.630> exist<00:05:51.389> so<00:05:52.389> why<00:05:53.199> does 00:05:53.399 --> 00:05:53.409 align:start position:0% it doesn't actually exist so why does 00:05:53.409 --> 00:05:55.379 align:start position:0% it doesn't actually exist so why does hardware<00:05:53.860> actually<00:05:54.099> matter<00:05:54.580> why<00:05:55.000> do<00:05:55.060> we<00:05:55.210> care 00:05:55.379 --> 00:05:55.389 align:start position:0% hardware actually matter why do we care 00:05:55.389 --> 00:05:59.100 align:start position:0% hardware actually matter why do we care about<00:05:55.449> hardware<00:05:56.050> with<00:05:56.229> aetherium<00:05:56.919> well<00:05:58.110> for 00:05:59.100 --> 00:05:59.110 align:start position:0% about hardware with aetherium well for 00:05:59.110 --> 00:06:01.050 align:start position:0% about hardware with aetherium well for the<00:05:59.199> time<00:05:59.380> being<00:05:59.680> hardware<00:06:00.310> actually<00:06:00.490> defines 00:06:01.050 --> 00:06:01.060 align:start position:0% the time being hardware actually defines 00:06:01.060 --> 00:06:04.409 align:start position:0% the time being hardware actually defines your<00:06:01.240> user<00:06:01.479> base<00:06:02.280> and<00:06:03.280> note<00:06:03.729> blockchain<00:06:04.300> is 00:06:04.409 --> 00:06:04.419 align:start position:0% your user base and note blockchain is 00:06:04.419 --> 00:06:07.320 align:start position:0% your user base and note blockchain is suited<00:06:04.900> for<00:06:04.930> all<00:06:05.169> user<00:06:05.590> bases<00:06:06.069> just<00:06:06.940> like<00:06:07.150> no 00:06:07.320 --> 00:06:07.330 align:start position:0% suited for all user bases just like no 00:06:07.330 --> 00:06:10.279 align:start position:0% suited for all user bases just like no application<00:06:07.930> is<00:06:08.050> suited<00:06:08.380> for<00:06:08.500> all<00:06:08.560> customers 00:06:10.279 --> 00:06:10.289 align:start position:0% application is suited for all customers 00:06:10.289 --> 00:06:13.980 align:start position:0% application is suited for all customers GPUs<00:06:11.460> therefore<00:06:12.460> the<00:06:12.520> common<00:06:12.819> folk<00:06:13.000> you<00:06:13.870> know 00:06:13.980 --> 00:06:13.990 align:start position:0% GPUs therefore the common folk you know 00:06:13.990 --> 00:06:15.719 align:start position:0% GPUs therefore the common folk you know they're<00:06:14.440> readily<00:06:14.740> available<00:06:14.830> they're 00:06:15.719 --> 00:06:15.729 align:start position:0% they're readily available they're 00:06:15.729 --> 00:06:17.850 align:start position:0% they're readily available they're flexible<00:06:16.150> they're<00:06:16.630> adaptable<00:06:17.080> they're<00:06:17.650> quite 00:06:17.850 --> 00:06:17.860 align:start position:0% flexible they're adaptable they're quite 00:06:17.860 --> 00:06:20.909 align:start position:0% flexible they're adaptable they're quite cheap<00:06:18.159> and<00:06:18.430> pretty<00:06:19.090> much<00:06:19.349> everyone<00:06:20.349> has<00:06:20.590> one 00:06:20.909 --> 00:06:20.919 align:start position:0% cheap and pretty much everyone has one 00:06:20.919 --> 00:06:23.399 align:start position:0% cheap and pretty much everyone has one today<00:06:21.370> look<00:06:22.180> at<00:06:22.270> your<00:06:22.389> phones<00:06:22.630> you<00:06:23.110> will<00:06:23.229> have 00:06:23.399 --> 00:06:23.409 align:start position:0% today look at your phones you will have 00:06:23.409 --> 00:06:25.680 align:start position:0% today look at your phones you will have a<00:06:23.440> GPU<00:06:23.800> in<00:06:24.009> there<00:06:24.190> every<00:06:24.909> single<00:06:25.210> laptop<00:06:25.389> you 00:06:25.680 --> 00:06:25.690 align:start position:0% a GPU in there every single laptop you 00:06:25.690 --> 00:06:28.860 align:start position:0% a GPU in there every single laptop you have<00:06:25.900> has<00:06:26.440> a<00:06:26.469> GPU<00:06:27.000> every<00:06:28.000> console<00:06:28.419> app<00:06:28.599> video 00:06:28.860 --> 00:06:28.870 align:start position:0% have has a GPU every console app video 00:06:28.870 --> 00:06:33.149 align:start position:0% have has a GPU every console app video game<00:06:29.199> console<00:06:29.620> out<00:06:29.710> there<00:06:29.889> has<00:06:30.130> a<00:06:30.159> GPU<00:06:32.009> FPGA<00:06:33.009> is 00:06:33.149 --> 00:06:33.159 align:start position:0% game console out there has a GPU FPGA is 00:06:33.159 --> 00:06:36.300 align:start position:0% game console out there has a GPU FPGA is there<00:06:33.789> for<00:06:34.000> enthusiasts<00:06:35.130> really<00:06:36.130> rich 00:06:36.300 --> 00:06:36.310 align:start position:0% there for enthusiasts really rich 00:06:36.310 --> 00:06:38.970 align:start position:0% there for enthusiasts really rich enthusiasts<00:06:37.120> they're<00:06:37.900> high<00:06:38.110> cost 00:06:38.970 --> 00:06:38.980 align:start position:0% enthusiasts they're high cost 00:06:38.980 --> 00:06:41.670 align:start position:0% enthusiasts they're high cost my<00:06:39.070> expertise<00:06:39.670> they<00:06:40.210> run<00:06:40.420> hot<00:06:40.660> the<00:06:41.320> definitely 00:06:41.670 --> 00:06:41.680 align:start position:0% my expertise they run hot the definitely 00:06:41.680 --> 00:06:43.860 align:start position:0% my expertise they run hot the definitely not<00:06:41.890> plug<00:06:42.100> and<00:06:42.250> play<00:06:42.430> and<00:06:42.700> I<00:06:43.270> should<00:06:43.480> know<00:06:43.660> the 00:06:43.860 --> 00:06:43.870 align:start position:0% not plug and play and I should know the 00:06:43.870 --> 00:06:45.930 align:start position:0% not plug and play and I should know the minority<00:06:44.320> group<00:06:44.500> hosts<00:06:44.890> around<00:06:45.040> 5,000<00:06:45.880> of 00:06:45.930 --> 00:06:45.940 align:start position:0% minority group hosts around 5,000 of 00:06:45.940 --> 00:06:46.590 align:start position:0% minority group hosts around 5,000 of these<00:06:46.060> things 00:06:46.590 --> 00:06:46.600 align:start position:0% these things 00:06:46.600 --> 00:06:51.000 align:start position:0% these things so<00:06:47.220> and<00:06:48.220> then<00:06:49.150> we<00:06:49.240> whoops<00:06:49.660> and<00:06:50.440> then<00:06:50.650> we<00:06:50.800> have 00:06:51.000 --> 00:06:51.010 align:start position:0% so and then we whoops and then we have 00:06:51.010 --> 00:06:54.480 align:start position:0% so and then we whoops and then we have of<00:06:51.520> course<00:06:51.690> Asics<00:06:52.690> these<00:06:52.960> are<00:06:53.740> really<00:06:54.220> for 00:06:54.480 --> 00:06:54.490 align:start position:0% of course Asics these are really for 00:06:54.490 --> 00:06:57.030 align:start position:0% of course Asics these are really for enterprise<00:06:55.090> minors<00:06:55.600> so<00:06:56.470> they're<00:06:56.590> relatively 00:06:57.030 --> 00:06:57.040 align:start position:0% enterprise minors so they're relatively 00:06:57.040 --> 00:06:58.800 align:start position:0% enterprise minors so they're relatively simple<00:06:57.580> they're<00:06:57.850> plug-and-play<00:06:58.150> they're 00:06:58.800 --> 00:06:58.810 align:start position:0% simple they're plug-and-play they're 00:06:58.810 --> 00:07:01.260 align:start position:0% simple they're plug-and-play they're consumable<00:06:59.470> goods<00:06:59.500> and<00:07:00.040> they<00:07:00.760> kind<00:07:01.000> of<00:07:01.090> have 00:07:01.260 --> 00:07:01.270 align:start position:0% consumable goods and they kind of have 00:07:01.270 --> 00:07:03.570 align:start position:0% consumable goods and they kind of have short<00:07:01.600> lifespans<00:07:02.170> and<00:07:02.530> they<00:07:02.980> generate<00:07:03.310> a<00:07:03.370> lot 00:07:03.570 --> 00:07:03.580 align:start position:0% short lifespans and they generate a lot 00:07:03.580 --> 00:07:06.150 align:start position:0% short lifespans and they generate a lot of<00:07:03.610> heat<00:07:03.910> and<00:07:04.180> consume<00:07:04.810> a<00:07:04.870> lot<00:07:05.020> of<00:07:05.050> power<00:07:05.320> but 00:07:06.150 --> 00:07:06.160 align:start position:0% of heat and consume a lot of power but 00:07:06.160 --> 00:07:08.010 align:start position:0% of heat and consume a lot of power but the<00:07:06.310> density<00:07:06.820> of<00:07:06.850> an<00:07:07.000> ASIC<00:07:07.390> lends<00:07:07.660> itself 00:07:08.010 --> 00:07:08.020 align:start position:0% the density of an ASIC lends itself 00:07:08.020 --> 00:07:13.110 align:start position:0% the density of an ASIC lends itself really<00:07:08.320> well<00:07:08.650> to<00:07:08.910> professional<00:07:09.910> farms<00:07:12.120> so 00:07:13.110 --> 00:07:13.120 align:start position:0% really well to professional farms so 00:07:13.120 --> 00:07:15.360 align:start position:0% really well to professional farms so what<00:07:13.330> user<00:07:13.630> base<00:07:13.870> do<00:07:14.200> we<00:07:14.320> actually<00:07:14.500> want<00:07:14.950> for 00:07:15.360 --> 00:07:15.370 align:start position:0% what user base do we actually want for 00:07:15.370 --> 00:07:19.170 align:start position:0% what user base do we actually want for aetherium<00:07:16.980> well<00:07:17.980> if<00:07:18.460> aetherium<00:07:19.000> is<00:07:19.150> a 00:07:19.170 --> 00:07:19.180 align:start position:0% aetherium well if aetherium is a 00:07:19.180 --> 00:07:21.150 align:start position:0% aetherium well if aetherium is a decentralized<00:07:19.930> application<00:07:20.620> blockchain 00:07:21.150 --> 00:07:21.160 align:start position:0% decentralized application blockchain 00:07:21.160 --> 00:07:23.070 align:start position:0% decentralized application blockchain that<00:07:21.910> means<00:07:22.150> we<00:07:22.330> kind<00:07:22.570> of<00:07:22.690> want<00:07:22.840> as<00:07:22.930> many 00:07:23.070 --> 00:07:23.080 align:start position:0% that means we kind of want as many 00:07:23.080 --> 00:07:24.990 align:start position:0% that means we kind of want as many participants<00:07:23.620> and<00:07:24.190> as<00:07:24.280> many<00:07:24.460> players<00:07:24.730> as 00:07:24.990 --> 00:07:25.000 align:start position:0% participants and as many players as 00:07:25.000 --> 00:07:28.620 align:start position:0% participants and as many players as possible<00:07:26.640> application<00:07:27.640> user<00:07:27.880> basis<00:07:28.390> usually 00:07:28.620 --> 00:07:28.630 align:start position:0% possible application user basis usually 00:07:28.630 --> 00:07:30.600 align:start position:0% possible application user basis usually need<00:07:29.140> to<00:07:29.380> be<00:07:29.500> as<00:07:29.650> decentralized<00:07:30.400> and 00:07:30.600 --> 00:07:30.610 align:start position:0% need to be as decentralized and 00:07:30.610 --> 00:07:32.490 align:start position:0% need to be as decentralized and distributed<00:07:30.640> as<00:07:31.420> possible<00:07:31.930> to<00:07:32.200> prevent 00:07:32.490 --> 00:07:32.500 align:start position:0% distributed as possible to prevent 00:07:32.500 --> 00:07:35.220 align:start position:0% distributed as possible to prevent malicious<00:07:33.070> actors<00:07:33.490> and<00:07:33.880> if<00:07:34.870> you<00:07:34.960> want<00:07:35.140> an 00:07:35.220 --> 00:07:35.230 align:start position:0% malicious actors and if you want an 00:07:35.230 --> 00:07:36.840 align:start position:0% malicious actors and if you want an enterprise-level<00:07:35.680> blockchain<00:07:36.580> you<00:07:36.700> also 00:07:36.840 --> 00:07:36.850 align:start position:0% enterprise-level blockchain you also 00:07:36.850 --> 00:07:38.460 align:start position:0% enterprise-level blockchain you also want<00:07:37.150> to<00:07:37.180> make<00:07:37.330> sure<00:07:37.360> as<00:07:37.630> many<00:07:37.810> participants 00:07:38.460 --> 00:07:38.470 align:start position:0% want to make sure as many participants 00:07:38.470 --> 00:07:41.340 align:start position:0% want to make sure as many participants as<00:07:38.860> possible<00:07:39.370> are<00:07:40.000> active<00:07:40.480> to<00:07:40.900> ensure<00:07:41.200> your 00:07:41.340 --> 00:07:41.350 align:start position:0% as possible are active to ensure your 00:07:41.350 --> 00:07:44.610 align:start position:0% as possible are active to ensure your application<00:07:41.490> your<00:07:42.490> code<00:07:42.790> is<00:07:43.350> executed<00:07:44.350> on<00:07:44.470> as 00:07:44.610 --> 00:07:44.620 align:start position:0% application your code is executed on as 00:07:44.620 --> 00:07:48.480 align:start position:0% application your code is executed on as many<00:07:44.860> diverse<00:07:45.100> nodes<00:07:45.610> as<00:07:45.760> possible<00:07:45.820> a<00:07:47.490> GPU 00:07:48.480 --> 00:07:48.490 align:start position:0% many diverse nodes as possible a GPU 00:07:48.490 --> 00:07:51.720 align:start position:0% many diverse nodes as possible a GPU card<00:07:48.820> is<00:07:49.000> naturally<00:07:49.470> decentralized<00:07:50.730> there's 00:07:51.720 --> 00:07:51.730 align:start position:0% card is naturally decentralized there's 00:07:51.730 --> 00:07:53.850 align:start position:0% card is naturally decentralized there's widespread<00:07:52.000> adoption<00:07:52.390> it's<00:07:53.170> in<00:07:53.320> every<00:07:53.560> device 00:07:53.850 --> 00:07:53.860 align:start position:0% widespread adoption it's in every device 00:07:53.860 --> 00:07:56.340 align:start position:0% widespread adoption it's in every device and<00:07:54.250> you<00:07:54.700> have<00:07:54.880> multiple<00:07:55.360> manufacturers<00:07:56.050> and 00:07:56.340 --> 00:07:56.350 align:start position:0% and you have multiple manufacturers and 00:07:56.350 --> 00:07:58.710 align:start position:0% and you have multiple manufacturers and a<00:07:56.410> healthy<00:07:56.710> ecosystem<00:07:57.330> outside<00:07:58.330> of<00:07:58.540> the 00:07:58.710 --> 00:07:58.720 align:start position:0% a healthy ecosystem outside of the 00:07:58.720 --> 00:08:01.800 align:start position:0% a healthy ecosystem outside of the blockchain<00:07:59.190> now<00:08:00.190> that<00:08:00.910> point<00:08:01.150> is<00:08:01.390> really 00:08:01.800 --> 00:08:01.810 align:start position:0% blockchain now that point is really 00:08:01.810 --> 00:08:04.200 align:start position:0% blockchain now that point is really important<00:08:02.400> because<00:08:03.400> when<00:08:03.700> you<00:08:03.790> only<00:08:03.940> have<00:08:04.180> one 00:08:04.200 --> 00:08:04.210 align:start position:0% important because when you only have one 00:08:04.210 --> 00:08:06.320 align:start position:0% important because when you only have one line<00:08:04.630> of<00:08:04.810> business<00:08:04.990> you<00:08:05.620> are<00:08:06.100> naturally 00:08:06.320 --> 00:08:06.330 align:start position:0% line of business you are naturally 00:08:06.330 --> 00:08:08.910 align:start position:0% line of business you are naturally incentivized<00:08:07.330> to<00:08:07.480> protect<00:08:07.870> it<00:08:07.990> at<00:08:08.080> all<00:08:08.230> costs 00:08:08.910 --> 00:08:08.920 align:start position:0% incentivized to protect it at all costs 00:08:08.920 --> 00:08:13.860 align:start position:0% incentivized to protect it at all costs that's<00:08:09.790> an<00:08:09.940> important<00:08:10.420> statement<00:08:12.690> but<00:08:13.690> wait 00:08:13.860 --> 00:08:13.870 align:start position:0% that's an important statement but wait 00:08:13.870 --> 00:08:18.690 align:start position:0% that's an important statement but wait what<00:08:14.680> about<00:08:15.300> CPUs<00:08:16.410> well<00:08:17.410> there's<00:08:18.220> a<00:08:18.310> problem 00:08:18.690 --> 00:08:18.700 align:start position:0% what about CPUs well there's a problem 00:08:18.700 --> 00:08:21.750 align:start position:0% what about CPUs well there's a problem with<00:08:19.000> CPUs<00:08:19.590> see<00:08:20.590> they're<00:08:21.310> naturally 00:08:21.750 --> 00:08:21.760 align:start position:0% with CPUs see they're naturally 00:08:21.760 --> 00:08:23.220 align:start position:0% with CPUs see they're naturally exploited<00:08:22.360> for<00:08:22.390> cryptocurrency<00:08:22.960> mining 00:08:23.220 --> 00:08:23.230 align:start position:0% exploited for cryptocurrency mining 00:08:23.230 --> 00:08:25.290 align:start position:0% exploited for cryptocurrency mining they're<00:08:23.980> really<00:08:24.190> great<00:08:24.400> with<00:08:24.520> botnets<00:08:25.000> and 00:08:25.290 --> 00:08:25.300 align:start position:0% they're really great with botnets and 00:08:25.300 --> 00:08:27.480 align:start position:0% they're really great with botnets and you<00:08:26.260> should<00:08:26.470> probably<00:08:26.770> trust<00:08:27.190> me<00:08:27.340> on<00:08:27.460> that 00:08:27.480 --> 00:08:27.490 align:start position:0% you should probably trust me on that 00:08:27.490 --> 00:08:29.790 align:start position:0% you should probably trust me on that because<00:08:28.060> I<00:08:28.420> might<00:08:29.020> have<00:08:29.140> written<00:08:29.290> a<00:08:29.470> few<00:08:29.620> back 00:08:29.790 --> 00:08:29.800 align:start position:0% because I might have written a few back 00:08:29.800 --> 00:08:31.250 align:start position:0% because I might have written a few back in<00:08:29.920> the<00:08:30.010> day 00:08:31.250 --> 00:08:31.260 align:start position:0% in the day 00:08:31.260 --> 00:08:34.020 align:start position:0% in the day CPU<00:08:32.260> implementations<00:08:33.099> are<00:08:33.250> also<00:08:33.400> the<00:08:33.669> easiest 00:08:34.020 --> 00:08:34.030 align:start position:0% CPU implementations are also the easiest 00:08:34.030 --> 00:08:36.440 align:start position:0% CPU implementations are also the easiest execute<00:08:34.690> on<00:08:34.840> fixed<00:08:35.200> function<00:08:35.440> Hardware 00:08:36.440 --> 00:08:36.450 align:start position:0% execute on fixed function Hardware 00:08:36.450 --> 00:08:38.580 align:start position:0% execute on fixed function Hardware because<00:08:37.450> of<00:08:37.570> their<00:08:37.690> very<00:08:37.870> single-threaded 00:08:38.580 --> 00:08:38.590 align:start position:0% because of their very single-threaded 00:08:38.590 --> 00:08:40.740 align:start position:0% because of their very single-threaded nature<00:08:39.070> that<00:08:39.580> lends<00:08:39.820> themself<00:08:40.240> really<00:08:40.599> well 00:08:40.740 --> 00:08:40.750 align:start position:0% nature that lends themself really well 00:08:40.750 --> 00:08:44.610 align:start position:0% nature that lends themself really well to<00:08:41.290> fix<00:08:41.530> silicon<00:08:42.130> implementations<00:08:42.910> on<00:08:43.470> top<00:08:44.470> of 00:08:44.610 --> 00:08:44.620 align:start position:0% to fix silicon implementations on top of 00:08:44.620 --> 00:08:47.400 align:start position:0% to fix silicon implementations on top of that<00:08:44.880> CPU<00:08:45.880> implementations<00:08:46.780> are<00:08:46.960> also<00:08:47.080> hard 00:08:47.400 --> 00:08:47.410 align:start position:0% that CPU implementations are also hard 00:08:47.410 --> 00:08:50.580 align:start position:0% that CPU implementations are also hard to<00:08:47.530> tune<00:08:47.770> for<00:08:48.010> each<00:08:48.850> architecture<00:08:49.720> shift<00:08:50.260> or 00:08:50.580 --> 00:08:50.590 align:start position:0% to tune for each architecture shift or 00:08:50.590 --> 00:08:52.680 align:start position:0% to tune for each architecture shift or each<00:08:50.710> revision<00:08:51.220> has<00:08:51.670> some<00:08:51.970> tricks<00:08:52.450> of 00:08:52.680 --> 00:08:52.690 align:start position:0% each revision has some tricks of 00:08:52.690 --> 00:08:56.310 align:start position:0% each revision has some tricks of trade<00:08:53.040> to<00:08:54.040> get<00:08:54.190> the<00:08:54.310> optimizations<00:08:55.090> or<00:08:55.420> you 00:08:56.310 --> 00:08:56.320 align:start position:0% trade to get the optimizations or you 00:08:56.320 --> 00:08:58.260 align:start position:0% trade to get the optimizations or you know<00:08:56.470> in<00:08:56.680> order<00:08:56.800> to<00:08:56.920> perfectly<00:08:57.700> saturate<00:08:58.180> the 00:08:58.260 --> 00:08:58.270 align:start position:0% know in order to perfectly saturate the 00:08:58.270 --> 00:09:01.260 align:start position:0% know in order to perfectly saturate the hardware<00:08:58.660> and<00:08:59.530> of<00:09:00.460> course<00:09:00.670> they're<00:09:01.000> not<00:09:01.150> that 00:09:01.260 --> 00:09:01.270 align:start position:0% hardware and of course they're not that 00:09:01.270 --> 00:09:03.780 align:start position:0% hardware and of course they're not that great<00:09:01.480> at<00:09:02.260> you<00:09:02.740> know<00:09:02.860> math<00:09:03.130> compared<00:09:03.610> to<00:09:03.730> an 00:09:03.780 --> 00:09:03.790 align:start position:0% great at you know math compared to an 00:09:03.790 --> 00:09:06.510 align:start position:0% great at you know math compared to an FPGA<00:09:04.270> or<00:09:04.510> a<00:09:04.540> GPU<00:09:04.930> they're<00:09:05.590> kind<00:09:05.770> of<00:09:05.860> slow<00:09:06.100> not 00:09:06.510 --> 00:09:06.520 align:start position:0% FPGA or a GPU they're kind of slow not 00:09:06.520 --> 00:09:09.240 align:start position:0% FPGA or a GPU they're kind of slow not as<00:09:06.640> dense<00:09:06.880> you<00:09:07.720> can't<00:09:07.990> really<00:09:08.380> pack<00:09:08.680> a<00:09:08.710> mining 00:09:09.240 --> 00:09:09.250 align:start position:0% as dense you can't really pack a mining 00:09:09.250 --> 00:09:11.730 align:start position:0% as dense you can't really pack a mining machine<00:09:09.580> full<00:09:09.790> of<00:09:09.940> CPUs<00:09:10.450> you<00:09:11.080> can<00:09:11.260> pack<00:09:11.410> a<00:09:11.440> rack 00:09:11.730 --> 00:09:11.740 align:start position:0% machine full of CPUs you can pack a rack 00:09:11.740 --> 00:09:14.760 align:start position:0% machine full of CPUs you can pack a rack full<00:09:12.010> of<00:09:12.040> Asics<00:09:12.460> or<00:09:12.610> GPUs<00:09:13.140> density<00:09:14.140> is<00:09:14.260> really 00:09:14.760 --> 00:09:14.770 align:start position:0% full of Asics or GPUs density is really 00:09:14.770 --> 00:09:16.830 align:start position:0% full of Asics or GPUs density is really really<00:09:14.980> important<00:09:15.550> to<00:09:15.700> mining<00:09:16.120> enthusiasts 00:09:16.830 --> 00:09:16.840 align:start position:0% really important to mining enthusiasts 00:09:16.840 --> 00:09:20.850 align:start position:0% really important to mining enthusiasts and<00:09:17.620> also<00:09:18.040> remember<00:09:18.550> GPU<00:09:19.150> was<00:09:19.750> a<00:09:20.080> GPU<00:09:20.620> based 00:09:20.850 --> 00:09:20.860 align:start position:0% and also remember GPU was a GPU based 00:09:20.860 --> 00:09:22.650 align:start position:0% and also remember GPU was a GPU based algorithm<00:09:21.490> from<00:09:21.670> the<00:09:21.790> start<00:09:22.120> that's<00:09:22.480> what 00:09:22.650 --> 00:09:22.660 align:start position:0% algorithm from the start that's what 00:09:22.660 --> 00:09:25.320 align:start position:0% algorithm from the start that's what memory<00:09:22.990> hardness<00:09:23.440> means<00:09:23.710> it<00:09:24.130> means<00:09:24.820> that<00:09:25.030> you 00:09:25.320 --> 00:09:25.330 align:start position:0% memory hardness means it means that you 00:09:25.330 --> 00:09:30.350 align:start position:0% memory hardness means it means that you need<00:09:25.540> a<00:09:25.810> GPU<00:09:26.320> we<00:09:27.160> simply<00:09:27.580> tuned<00:09:28.090> it<00:09:28.270> for<00:09:28.540> GPUs 00:09:30.350 --> 00:09:30.360 align:start position:0% need a GPU we simply tuned it for GPUs 00:09:30.360 --> 00:09:33.180 align:start position:0% need a GPU we simply tuned it for GPUs now<00:09:31.360> a<00:09:31.660> warning<00:09:32.200> this<00:09:32.470> is<00:09:32.590> gonna<00:09:32.710> get<00:09:32.950> a<00:09:33.010> bit 00:09:33.180 --> 00:09:33.190 align:start position:0% now a warning this is gonna get a bit 00:09:33.190 --> 00:09:35.430 align:start position:0% now a warning this is gonna get a bit technical<00:09:33.430> from<00:09:33.820> this<00:09:33.970> point<00:09:34.180> on<00:09:34.330> this<00:09:35.110> is<00:09:35.290> our 00:09:35.430 --> 00:09:35.440 align:start position:0% technical from this point on this is our 00:09:35.440 --> 00:09:38.640 align:start position:0% technical from this point on this is our at<00:09:35.620> hash<00:09:35.860> spaghetti<00:09:36.310> and<00:09:36.490> logo<00:09:37.170> Lego<00:09:38.170> rather 00:09:38.640 --> 00:09:38.650 align:start position:0% at hash spaghetti and logo Lego rather 00:09:38.650 --> 00:09:41.640 align:start position:0% at hash spaghetti and logo Lego rather and<00:09:38.980> hash<00:09:39.400> itself<00:09:39.820> is<00:09:40.060> relatively<00:09:40.660> simple<00:09:41.080> it 00:09:41.640 --> 00:09:41.650 align:start position:0% and hash itself is relatively simple it 00:09:41.650 --> 00:09:44.070 align:start position:0% and hash itself is relatively simple it requires<00:09:42.220> a<00:09:42.340> semi<00:09:42.730> constant<00:09:43.330> scratch<00:09:43.570> pad 00:09:44.070 --> 00:09:44.080 align:start position:0% requires a semi constant scratch pad 00:09:44.080 --> 00:09:46.230 align:start position:0% requires a semi constant scratch pad file<00:09:44.350> we<00:09:44.590> call<00:09:44.770> it<00:09:44.860> the<00:09:44.950> dag<00:09:45.160> and<00:09:45.820> then<00:09:46.060> all 00:09:46.230 --> 00:09:46.240 align:start position:0% file we call it the dag and then all 00:09:46.240 --> 00:09:48.870 align:start position:0% file we call it the dag and then all instances<00:09:46.780> reuse<00:09:47.170> that<00:09:47.350> same<00:09:47.410> dag<00:09:47.950> for<00:09:48.790> the 00:09:48.870 --> 00:09:48.880 align:start position:0% instances reuse that same dag for the 00:09:48.880 --> 00:09:52.800 align:start position:0% instances reuse that same dag for the next<00:09:49.120> 100<00:09:49.660> hours<00:09:50.820> mining<00:09:51.820> it<00:09:51.970> hash<00:09:52.180> really 00:09:52.800 --> 00:09:52.810 align:start position:0% next 100 hours mining it hash really 00:09:52.810 --> 00:09:54.900 align:start position:0% next 100 hours mining it hash really just<00:09:52.990> involves<00:09:53.350> grabbing<00:09:53.710> random<00:09:54.100> slices<00:09:54.760> of 00:09:54.900 --> 00:09:54.910 align:start position:0% just involves grabbing random slices of 00:09:54.910 --> 00:09:56.160 align:start position:0% just involves grabbing random slices of the<00:09:54.970> data<00:09:55.000> set<00:09:55.420> and<00:09:55.540> hashing<00:09:55.900> them<00:09:56.050> together 00:09:56.160 --> 00:09:56.170 align:start position:0% the data set and hashing them together 00:09:56.170 --> 00:09:58.410 align:start position:0% the data set and hashing them together and<00:09:56.710> then<00:09:56.860> the<00:09:56.980> DRAM<00:09:57.340> peak<00:09:57.790> bandwidth<00:09:58.240> up 00:09:58.410 --> 00:09:58.420 align:start position:0% and then the DRAM peak bandwidth up 00:09:58.420 --> 00:10:00.420 align:start position:0% and then the DRAM peak bandwidth up there<00:09:58.600> which<00:09:58.780> we<00:09:58.930> can<00:09:59.080> see<00:09:59.260> it's<00:10:00.010> easy<00:10:00.250> to 00:10:00.420 --> 00:10:00.430 align:start position:0% there which we can see it's easy to 00:10:00.430 --> 00:10:03.240 align:start position:0% there which we can see it's easy to calculate<00:10:01.060> it's<00:10:01.720> 64<00:10:02.260> iterations<00:10:02.920> each 00:10:03.240 --> 00:10:03.250 align:start position:0% calculate it's 64 iterations each 00:10:03.250 --> 00:10:07.620 align:start position:0% calculate it's 64 iterations each reading<00:10:03.550> 128<00:10:04.360> sequential<00:10:04.930> bytes<00:10:05.910> so<00:10:06.910> at<00:10:07.330> hash 00:10:07.620 --> 00:10:07.630 align:start position:0% reading 128 sequential bytes so at hash 00:10:07.630 --> 00:10:11.460 align:start position:0% reading 128 sequential bytes so at hash is<00:10:07.990> really<00:10:08.680> just<00:10:09.100> a<00:10:09.330> memory<00:10:10.330> hard<00:10:10.870> proof<00:10:11.260> of 00:10:11.460 --> 00:10:11.470 align:start position:0% is really just a memory hard proof of 00:10:11.470 --> 00:10:14.910 align:start position:0% is really just a memory hard proof of work<00:10:11.620> algorithm<00:10:12.690> that<00:10:13.690> means<00:10:13.930> it<00:10:14.620> just 00:10:14.910 --> 00:10:14.920 align:start position:0% work algorithm that means it just 00:10:14.920 --> 00:10:17.460 align:start position:0% work algorithm that means it just requires<00:10:15.250> two<00:10:15.760> major<00:10:16.120> things<00:10:16.480> a<00:10:16.810> relatively 00:10:17.460 --> 00:10:17.470 align:start position:0% requires two major things a relatively 00:10:17.470 --> 00:10:19.800 align:start position:0% requires two major things a relatively large<00:10:17.770> frame<00:10:18.070> buffer<00:10:18.460> and<00:10:18.670> as<00:10:19.330> much<00:10:19.510> memory 00:10:19.800 --> 00:10:19.810 align:start position:0% large frame buffer and as much memory 00:10:19.810 --> 00:10:22.260 align:start position:0% large frame buffer and as much memory bandwidth<00:10:20.290> as<00:10:20.470> possible<00:10:20.920> both<00:10:21.820> of<00:10:22.090> these 00:10:22.260 --> 00:10:22.270 align:start position:0% bandwidth as possible both of these 00:10:22.270 --> 00:10:25.410 align:start position:0% bandwidth as possible both of these things<00:10:22.690> that<00:10:22.870> GPU<00:10:23.260> has<00:10:23.500> great<00:10:24.400> but<00:10:24.970> guess<00:10:25.150> what 00:10:25.410 --> 00:10:25.420 align:start position:0% things that GPU has great but guess what 00:10:25.420 --> 00:10:28.800 align:start position:0% things that GPU has great but guess what an<00:10:25.600> ASIC<00:10:26.020> can<00:10:26.170> have<00:10:26.350> that<00:10:26.590> too<00:10:27.030> so<00:10:28.030> can<00:10:28.210> an<00:10:28.300> FPGA 00:10:28.800 --> 00:10:28.810 align:start position:0% an ASIC can have that too so can an FPGA 00:10:28.810 --> 00:10:31.590 align:start position:0% an ASIC can have that too so can an FPGA so<00:10:29.770> it<00:10:29.860> can<00:10:29.980> a<00:10:30.010> beat<00:10:30.220> PCB<00:10:30.880> with<00:10:31.090> just<00:10:31.330> a<00:10:31.420> ton<00:10:31.570> of 00:10:31.590 --> 00:10:31.600 align:start position:0% so it can a beat PCB with just a ton of 00:10:31.600 --> 00:10:33.810 align:start position:0% so it can a beat PCB with just a ton of DRAM<00:10:32.020> in<00:10:32.200> a<00:10:32.260> little<00:10:32.440> compute<00:10:32.770> core<00:10:32.980> because 00:10:33.810 --> 00:10:33.820 align:start position:0% DRAM in a little compute core because 00:10:33.820 --> 00:10:36.090 align:start position:0% DRAM in a little compute core because that's<00:10:34.000> all<00:10:34.180> you<00:10:34.330> need<00:10:34.510> memory<00:10:35.050> DRAM<00:10:35.589> chips 00:10:36.090 --> 00:10:36.100 align:start position:0% that's all you need memory DRAM chips 00:10:36.100 --> 00:10:38.790 align:start position:0% that's all you need memory DRAM chips and<00:10:36.520> hashes<00:10:37.480> only<00:10:37.810> tuned<00:10:38.110> to<00:10:38.230> memory<00:10:38.589> but<00:10:38.740> not 00:10:38.790 --> 00:10:38.800 align:start position:0% and hashes only tuned to memory but not 00:10:38.800 --> 00:10:41.100 align:start position:0% and hashes only tuned to memory but not the<00:10:39.100> rest<00:10:39.280> of<00:10:39.430> the<00:10:39.520> hardware<00:10:40.000> and<00:10:40.210> so<00:10:40.600> again 00:10:41.100 --> 00:10:41.110 align:start position:0% the rest of the hardware and so again 00:10:41.110 --> 00:10:43.770 align:start position:0% the rest of the hardware and so again how<00:10:42.100> you<00:10:42.160> make<00:10:42.430> an<00:10:42.550> ASIC<00:10:42.910> you<00:10:43.120> just<00:10:43.300> remove<00:10:43.570> all 00:10:43.770 --> 00:10:43.780 align:start position:0% how you make an ASIC you just remove all 00:10:43.780 --> 00:10:45.090 align:start position:0% how you make an ASIC you just remove all the<00:10:43.930> other<00:10:44.080> parts<00:10:44.500> that<00:10:44.560> are<00:10:44.770> currently 00:10:45.090 --> 00:10:45.100 align:start position:0% the other parts that are currently 00:10:45.100 --> 00:10:47.460 align:start position:0% the other parts that are currently unused<00:10:45.280> or<00:10:45.880> rather<00:10:46.420> all<00:10:46.570> the<00:10:46.870> wasted<00:10:47.230> space 00:10:47.460 --> 00:10:47.470 align:start position:0% unused or rather all the wasted space 00:10:47.470 --> 00:10:50.940 align:start position:0% unused or rather all the wasted space and<00:10:47.860> you<00:10:48.310> get<00:10:48.460> a<00:10:48.490> nice<00:10:48.760> increase<00:10:49.530> that's<00:10:50.530> ASIC 00:10:50.940 --> 00:10:50.950 align:start position:0% and you get a nice increase that's ASIC 00:10:50.950 --> 00:10:53.520 align:start position:0% and you get a nice increase that's ASIC creation<00:10:51.339> in<00:10:51.460> a<00:10:51.550> nutshell<00:10:51.730> guys<00:10:52.260> just<00:10:53.260> remove 00:10:53.520 --> 00:10:53.530 align:start position:0% creation in a nutshell guys just remove 00:10:53.530 --> 00:10:56.579 align:start position:0% creation in a nutshell guys just remove all<00:10:53.680> the<00:10:53.830> unwanted<00:10:54.460> or<00:10:54.910> wasted<00:10:55.570> parts<00:10:56.110> be<00:10:56.500> it 00:10:56.579 --> 00:10:56.589 align:start position:0% all the unwanted or wasted parts be it 00:10:56.589 --> 00:10:58.020 align:start position:0% all the unwanted or wasted parts be it silicon<00:10:57.010> or<00:10:57.160> compute<00:10:57.520> cycles 00:10:58.020 --> 00:10:58.030 align:start position:0% silicon or compute cycles 00:10:58.030 --> 00:11:02.880 align:start position:0% silicon or compute cycles voila<00:11:00.600> so<00:11:01.600> the<00:11:01.720> inefficiencies<00:11:01.960> in<00:11:02.440> add<00:11:02.680> hash 00:11:02.880 --> 00:11:02.890 align:start position:0% voila so the inefficiencies in add hash 00:11:02.890 --> 00:11:04.500 align:start position:0% voila so the inefficiencies in add hash become<00:11:03.280> more<00:11:03.460> obvious<00:11:03.580> when<00:11:03.970> you<00:11:04.030> profile<00:11:04.480> a 00:11:04.500 --> 00:11:04.510 align:start position:0% become more obvious when you profile a 00:11:04.510 --> 00:11:05.630 align:start position:0% become more obvious when you profile a graphics<00:11:04.930> card<00:11:05.170> with 00:11:05.630 --> 00:11:05.640 align:start position:0% graphics card with 00:11:05.640 --> 00:11:08.870 align:start position:0% graphics card with if<00:11:05.970> minor<00:11:06.329> so<00:11:07.230> this<00:11:07.260> is<00:11:07.529> the<00:11:08.010> top<00:11:08.640> picture 00:11:08.870 --> 00:11:08.880 align:start position:0% if minor so this is the top picture 00:11:08.880 --> 00:11:10.910 align:start position:0% if minor so this is the top picture there<00:11:09.180> is<00:11:09.209> taken<00:11:09.660> from<00:11:09.810> insight<00:11:10.410> profiler 00:11:10.910 --> 00:11:10.920 align:start position:0% there is taken from insight profiler 00:11:10.920 --> 00:11:12.530 align:start position:0% there is taken from insight profiler that's<00:11:11.339> new<00:11:11.550> videos<00:11:11.850> new<00:11:11.880> performance 00:11:12.530 --> 00:11:12.540 align:start position:0% that's new videos new performance 00:11:12.540 --> 00:11:15.790 align:start position:0% that's new videos new performance profiler<00:11:12.959> pretty<00:11:13.529> sexy<00:11:13.950> bottom<00:11:14.670> one<00:11:14.820> is<00:11:14.970> AMD's 00:11:15.790 --> 00:11:15.800 align:start position:0% profiler pretty sexy bottom one is AMD's 00:11:15.800 --> 00:11:19.730 align:start position:0% profiler pretty sexy bottom one is AMD's so<00:11:16.800> you<00:11:17.250> see<00:11:17.459> SM<00:11:17.850> up<00:11:18.029> there<00:11:18.209> s<00:11:18.690> M<00:11:19.350> stands<00:11:19.620> for 00:11:19.730 --> 00:11:19.740 align:start position:0% so you see SM up there s M stands for 00:11:19.740 --> 00:11:22.160 align:start position:0% so you see SM up there s M stands for streaming<00:11:20.190> multi<00:11:20.579> presser<00:11:20.940> multi<00:11:21.630> processor 00:11:22.160 --> 00:11:22.170 align:start position:0% streaming multi presser multi processor 00:11:22.170 --> 00:11:23.900 align:start position:0% streaming multi presser multi processor those<00:11:22.680> are<00:11:22.860> kind<00:11:23.010> of<00:11:23.070> the<00:11:23.160> computational 00:11:23.900 --> 00:11:23.910 align:start position:0% those are kind of the computational 00:11:23.910 --> 00:11:26.329 align:start position:0% those are kind of the computational cores<00:11:24.180> of<00:11:24.329> NVIDIA<00:11:24.839> GPUs<00:11:25.320> which<00:11:26.070> you<00:11:26.220> know 00:11:26.329 --> 00:11:26.339 align:start position:0% cores of NVIDIA GPUs which you know 00:11:26.339 --> 00:11:28.490 align:start position:0% cores of NVIDIA GPUs which you know create<00:11:26.700> manage<00:11:27.240> schedule<00:11:27.839> execute 00:11:28.490 --> 00:11:28.500 align:start position:0% create manage schedule execute 00:11:28.500 --> 00:11:31.040 align:start position:0% create manage schedule execute instructions<00:11:29.329> for<00:11:30.329> many<00:11:30.510> threads<00:11:30.839> in 00:11:31.040 --> 00:11:31.050 align:start position:0% instructions for many threads in 00:11:31.050 --> 00:11:34.100 align:start position:0% instructions for many threads in parallel<00:11:31.079> and<00:11:31.980> they<00:11:32.970> consume<00:11:33.480> most<00:11:33.870> of<00:11:34.050> the 00:11:34.100 --> 00:11:34.110 align:start position:0% parallel and they consume most of the 00:11:34.110 --> 00:11:36.319 align:start position:0% parallel and they consume most of the GPUs<00:11:34.529> die<00:11:34.800> area<00:11:35.220> so<00:11:35.640> when<00:11:35.790> we<00:11:35.880> talk<00:11:36.060> about<00:11:36.089> a 00:11:36.319 --> 00:11:36.329 align:start position:0% GPUs die area so when we talk about a 00:11:36.329 --> 00:11:39.250 align:start position:0% GPUs die area so when we talk about a GPU<00:11:36.720> ASIC<00:11:37.230> we're<00:11:37.440> referring<00:11:37.709> to<00:11:37.829> the<00:11:37.980> GPU<00:11:38.310> core 00:11:39.250 --> 00:11:39.260 align:start position:0% GPU ASIC we're referring to the GPU core 00:11:39.260 --> 00:11:42.350 align:start position:0% GPU ASIC we're referring to the GPU core as<00:11:40.260> we<00:11:40.500> can<00:11:40.649> see<00:11:40.829> they<00:11:41.070> run<00:11:41.279> up<00:11:41.430> less<00:11:41.670> than<00:11:41.700> 30% 00:11:42.350 --> 00:11:42.360 align:start position:0% as we can see they run up less than 30% 00:11:42.360 --> 00:11:45.019 align:start position:0% as we can see they run up less than 30% utilization<00:11:42.839> right<00:11:43.560> now<00:11:43.740> and<00:11:44.040> all<00:11:44.579> the<00:11:44.730> bottom 00:11:45.019 --> 00:11:45.029 align:start position:0% utilization right now and all the bottom 00:11:45.029 --> 00:11:48.560 align:start position:0% utilization right now and all the bottom code<00:11:45.570> Excel<00:11:45.959> AMD's<00:11:46.649> performance<00:11:47.399> profiler<00:11:47.880> we 00:11:48.560 --> 00:11:48.570 align:start position:0% code Excel AMD's performance profiler we 00:11:48.570 --> 00:11:50.389 align:start position:0% code Excel AMD's performance profiler we don't<00:11:48.779> have<00:11:49.110> as<00:11:49.320> many<00:11:49.440> details<00:11:49.740> but<00:11:50.160> you<00:11:50.279> can 00:11:50.389 --> 00:11:50.399 align:start position:0% don't have as many details but you can 00:11:50.399 --> 00:11:53.660 align:start position:0% don't have as many details but you can see<00:11:50.610> there<00:11:50.959> v<00:11:51.959> al<00:11:52.560> you<00:11:52.920> busy 00:11:53.660 --> 00:11:53.670 align:start position:0% see there v al you busy 00:11:53.670 --> 00:11:57.110 align:start position:0% see there v al you busy that<00:11:54.269> stands<00:11:54.570> for<00:11:54.660> vector<00:11:55.050> al<00:11:55.350> use<00:11:55.820> vector<00:11:56.820> al 00:11:57.110 --> 00:11:57.120 align:start position:0% that stands for vector al use vector al 00:11:57.120 --> 00:11:58.670 align:start position:0% that stands for vector al use vector al use<00:11:57.390> are<00:11:57.600> the<00:11:57.720> digital<00:11:58.050> circuits<00:11:58.500> that 00:11:58.670 --> 00:11:58.680 align:start position:0% use are the digital circuits that 00:11:58.680 --> 00:12:00.259 align:start position:0% use are the digital circuits that perform<00:11:59.040> arithmetic<00:11:59.279> and<00:11:59.880> bitwise 00:12:00.259 --> 00:12:00.269 align:start position:0% perform arithmetic and bitwise 00:12:00.269 --> 00:12:03.530 align:start position:0% perform arithmetic and bitwise operations<00:12:01.050> for<00:12:01.260> the<00:12:01.350> AMG<00:12:01.620> AMD<00:12:02.519> GPU<00:12:02.940> cards<00:12:03.300> and 00:12:03.530 --> 00:12:03.540 align:start position:0% operations for the AMG AMD GPU cards and 00:12:03.540 --> 00:12:05.810 align:start position:0% operations for the AMG AMD GPU cards and it's<00:12:04.200> kind<00:12:04.410> of<00:12:04.470> the<00:12:04.560> same<00:12:04.769> thing<00:12:04.980> as<00:12:05.160> math<00:12:05.550> core 00:12:05.810 --> 00:12:05.820 align:start position:0% it's kind of the same thing as math core 00:12:05.820 --> 00:12:09.110 align:start position:0% it's kind of the same thing as math core for<00:12:06.570> NVIDIA<00:12:06.750> GPUs<00:12:07.380> both<00:12:08.190> of<00:12:08.399> these<00:12:08.490> things<00:12:08.790> are 00:12:09.110 --> 00:12:09.120 align:start position:0% for NVIDIA GPUs both of these things are 00:12:09.120 --> 00:12:11.720 align:start position:0% for NVIDIA GPUs both of these things are the<00:12:09.300> building<00:12:09.630> blocks<00:12:09.839> of<00:12:09.959> a<00:12:10.050> GPU<00:12:10.380> ASIC<00:12:10.829> so 00:12:11.720 --> 00:12:11.730 align:start position:0% the building blocks of a GPU ASIC so 00:12:11.730 --> 00:12:14.740 align:start position:0% the building blocks of a GPU ASIC so that<00:12:11.880> means<00:12:12.120> today<00:12:12.420> at<00:12:12.779> hash<00:12:13.170> only<00:12:13.800> consumes 00:12:14.740 --> 00:12:14.750 align:start position:0% that means today at hash only consumes 00:12:14.750 --> 00:12:20.540 align:start position:0% that means today at hash only consumes 30%<00:12:15.750> less<00:12:16.050> than<00:12:16.110> 30%<00:12:16.620> of<00:12:17.360> a<00:12:18.360> GPU<00:12:18.690> core<00:12:19.550> so 00:12:20.540 --> 00:12:20.550 align:start position:0% 30% less than 30% of a GPU core so 00:12:20.550 --> 00:12:22.610 align:start position:0% 30% less than 30% of a GPU core so that's<00:12:20.730> one<00:12:21.360> of<00:12:21.480> the<00:12:21.570> major<00:12:21.810> deficiencies 00:12:22.610 --> 00:12:22.620 align:start position:0% that's one of the major deficiencies 00:12:22.620 --> 00:12:26.930 align:start position:0% that's one of the major deficiencies with<00:12:23.040> it<00:12:23.220> hash<00:12:23.459> today<00:12:24.470> and<00:12:25.490> the<00:12:26.490> reason<00:12:26.820> you 00:12:26.930 --> 00:12:26.940 align:start position:0% with it hash today and the reason you 00:12:26.940 --> 00:12:29.329 align:start position:0% with it hash today and the reason you can<00:12:27.089> get<00:12:27.209> a<00:12:27.390> large<00:12:27.750> ASIC<00:12:28.649> performance<00:12:29.160> gained 00:12:29.329 --> 00:12:29.339 align:start position:0% can get a large ASIC performance gained 00:12:29.339 --> 00:12:31.670 align:start position:0% can get a large ASIC performance gained because<00:12:29.760> really<00:12:30.449> all<00:12:30.570> it<00:12:30.870> does<00:12:31.050> is<00:12:31.320> a<00:12:31.350> small 00:12:31.670 --> 00:12:31.680 align:start position:0% because really all it does is a small 00:12:31.680 --> 00:12:35.509 align:start position:0% because really all it does is a small 128<00:12:32.610> byte<00:12:33.269> read<00:12:33.540> from<00:12:33.810> the<00:12:33.899> main<00:12:34.079> memory<00:12:34.519> that 00:12:35.509 --> 00:12:35.519 align:start position:0% 128 byte read from the main memory that 00:12:35.519 --> 00:12:38.240 align:start position:0% 128 byte read from the main memory that small<00:12:36.029> access<00:12:36.570> size<00:12:36.870> is<00:12:37.140> a<00:12:37.170> reason<00:12:37.620> that<00:12:37.649> GPUs 00:12:38.240 --> 00:12:38.250 align:start position:0% small access size is a reason that GPUs 00:12:38.250 --> 00:12:42.319 align:start position:0% small access size is a reason that GPUs that<00:12:38.279> utilize<00:12:38.850> gddr5<00:12:39.690> X<00:12:40.290> or<00:12:40.769> GD<00:12:41.190> DDR<00:12:41.550> 6<00:12:41.910> memory 00:12:42.319 --> 00:12:42.329 align:start position:0% that utilize gddr5 X or GD DDR 6 memory 00:12:42.329 --> 00:12:45.220 align:start position:0% that utilize gddr5 X or GD DDR 6 memory were<00:12:42.779> so<00:12:43.019> inefficient<00:12:43.500> at<00:12:43.800> executing<00:12:44.399> at<00:12:44.519> hash 00:12:45.220 --> 00:12:45.230 align:start position:0% were so inefficient at executing at hash 00:12:45.230 --> 00:12:47.630 align:start position:0% were so inefficient at executing at hash some<00:12:46.230> of<00:12:46.350> you<00:12:46.470> may<00:12:46.620> be<00:12:46.649> familiar<00:12:47.010> with<00:12:47.339> our 00:12:47.630 --> 00:12:47.640 align:start position:0% some of you may be familiar with our 00:12:47.640 --> 00:12:50.630 align:start position:0% some of you may be familiar with our tool<00:12:48.240> f<00:12:48.540> largeman<00:12:49.050> it<00:12:49.290> got<00:12:49.829> kind<00:12:50.100> of<00:12:50.190> viral<00:12:50.550> on 00:12:50.630 --> 00:12:50.640 align:start position:0% tool f largeman it got kind of viral on 00:12:50.640 --> 00:12:54.769 align:start position:0% tool f largeman it got kind of viral on the<00:12:50.670> internet<00:12:51.709> watchmen<00:12:52.709> abused<00:12:53.130> this<00:12:53.570> so<00:12:54.570> we 00:12:54.769 --> 00:12:54.779 align:start position:0% the internet watchmen abused this so we 00:12:54.779 --> 00:12:57.769 align:start position:0% the internet watchmen abused this so we matched<00:12:55.050> access<00:12:55.620> patterns<00:12:56.070> of<00:12:56.339> 5x<00:12:57.060> and<00:12:57.390> g<00:12:57.540> 6 00:12:57.769 --> 00:12:57.779 align:start position:0% matched access patterns of 5x and g 6 00:12:57.779 --> 00:13:00.740 align:start position:0% matched access patterns of 5x and g 6 memory<00:12:58.199> to<00:12:58.680> it<00:12:58.829> hash<00:12:59.100> to<00:12:59.399> enable<00:12:59.790> 128<00:13:00.510> byte 00:13:00.740 --> 00:13:00.750 align:start position:0% memory to it hash to enable 128 byte 00:13:00.750 --> 00:13:03.590 align:start position:0% memory to it hash to enable 128 byte loads<00:13:01.500> to<00:13:01.829> run<00:13:02.010> at<00:13:02.130> full<00:13:02.279> speed<00:13:02.670> we<00:13:03.420> have<00:13:03.449> a 00:13:03.590 --> 00:13:03.600 align:start position:0% loads to run at full speed we have a 00:13:03.600 --> 00:13:05.840 align:start position:0% loads to run at full speed we have a public<00:13:04.050> version<00:13:04.230> of<00:13:04.470> that<00:13:04.589> tool<00:13:04.860> for<00:13:05.160> 5x<00:13:05.579> we 00:13:05.840 --> 00:13:05.850 align:start position:0% public version of that tool for 5x we 00:13:05.850 --> 00:13:07.490 align:start position:0% public version of that tool for 5x we have<00:13:05.970> a<00:13:06.000> private<00:13:06.240> version<00:13:06.540> of<00:13:06.779> that<00:13:06.899> tool<00:13:07.170> for 00:13:07.490 --> 00:13:07.500 align:start position:0% have a private version of that tool for 00:13:07.500 --> 00:13:11.150 align:start position:0% have a private version of that tool for G<00:13:08.130> 6<00:13:08.480> we<00:13:09.480> made<00:13:09.630> a<00:13:09.660> lot<00:13:09.899> of<00:13:09.990> money<00:13:10.230> off<00:13:10.410> that<00:13:10.649> just 00:13:11.150 --> 00:13:11.160 align:start position:0% G 6 we made a lot of money off that just 00:13:11.160 --> 00:13:13.400 align:start position:0% G 6 we made a lot of money off that just because<00:13:11.459> an<00:13:11.550> algorithm<00:13:12.089> wasn't<00:13:12.480> effectively 00:13:13.400 --> 00:13:13.410 align:start position:0% because an algorithm wasn't effectively 00:13:13.410 --> 00:13:16.970 align:start position:0% because an algorithm wasn't effectively tuned<00:13:13.709> to<00:13:14.399> the<00:13:14.699> hardware<00:13:15.410> we<00:13:16.410> see<00:13:16.620> up<00:13:16.740> there 00:13:16.970 --> 00:13:16.980 align:start position:0% tuned to the hardware we see up there 00:13:16.980 --> 00:13:18.660 align:start position:0% tuned to the hardware we see up there the<00:13:17.279> tide<00:13:17.520> and<00:13:17.670> X<00:13:17.880> pass 00:13:18.660 --> 00:13:18.670 align:start position:0% the tide and X pass 00:13:18.670 --> 00:13:21.090 align:start position:0% the tide and X pass it's<00:13:19.060> similar<00:13:19.300> to<00:13:19.450> a<00:13:19.600> 1080<00:13:20.020> TI<00:13:20.320> and<00:13:20.890> it's 00:13:21.090 --> 00:13:21.100 align:start position:0% it's similar to a 1080 TI and it's 00:13:21.100 --> 00:13:23.040 align:start position:0% it's similar to a 1080 TI and it's pretty<00:13:21.430> awful<00:13:21.580> where<00:13:22.000> you<00:13:22.480> can<00:13:22.630> see<00:13:22.810> up<00:13:22.930> there 00:13:23.040 --> 00:13:23.050 align:start position:0% pretty awful where you can see up there 00:13:23.050 --> 00:13:25.350 align:start position:0% pretty awful where you can see up there it<00:13:23.200> has<00:13:23.350> less<00:13:23.589> than<00:13:23.770> 20%<00:13:24.040> core<00:13:24.550> utilization 00:13:25.350 --> 00:13:25.360 align:start position:0% it has less than 20% core utilization 00:13:25.360 --> 00:13:27.720 align:start position:0% it has less than 20% core utilization and<00:13:25.570> less<00:13:26.050> than<00:13:26.260> 60%<00:13:26.800> memory<00:13:27.550> utilization 00:13:27.720 --> 00:13:27.730 align:start position:0% and less than 60% memory utilization 00:13:27.730 --> 00:13:29.570 align:start position:0% and less than 60% memory utilization which<00:13:28.480> is<00:13:28.630> on<00:13:28.750> the<00:13:28.839> bottom 00:13:29.570 --> 00:13:29.580 align:start position:0% which is on the bottom 00:13:29.580 --> 00:13:33.630 align:start position:0% which is on the bottom there's<00:13:30.580> also<00:13:30.850> another<00:13:31.420> issue<00:13:32.370> cat<00:13:33.370> jack 00:13:33.630 --> 00:13:33.640 align:start position:0% there's also another issue cat jack 00:13:33.640 --> 00:13:35.700 align:start position:0% there's also another issue cat jack that's<00:13:34.390> the<00:13:34.540> hash<00:13:34.690> function<00:13:34.930> that<00:13:35.260> stuff<00:13:35.500> at 00:13:35.700 --> 00:13:35.710 align:start position:0% that's the hash function that stuff at 00:13:35.710 --> 00:13:38.519 align:start position:0% that's the hash function that stuff at the<00:13:35.890> start<00:13:36.130> and<00:13:36.279> end<00:13:36.460> of<00:13:36.610> it<00:13:36.760> hash<00:13:36.940> and<00:13:37.529> that 00:13:38.519 --> 00:13:38.529 align:start position:0% the start and end of it hash and that 00:13:38.529 --> 00:13:41.130 align:start position:0% the start and end of it hash and that can<00:13:38.800> be<00:13:38.950> executed<00:13:39.730> much<00:13:40.089> more<00:13:40.450> efficiently<00:13:41.020> on 00:13:41.130 --> 00:13:41.140 align:start position:0% can be executed much more efficiently on 00:13:41.140 --> 00:13:45.210 align:start position:0% can be executed much more efficiently on an<00:13:41.440> FPGA<00:13:42.040> or<00:13:42.160> an<00:13:42.400> ASIC<00:13:43.089> in<00:13:43.410> fact<00:13:44.410> that's 00:13:45.210 --> 00:13:45.220 align:start position:0% an FPGA or an ASIC in fact that's 00:13:45.220 --> 00:13:48.870 align:start position:0% an FPGA or an ASIC in fact that's actually<00:13:45.790> what<00:13:46.690> the<00:13:47.350> acorn<00:13:47.860> line<00:13:48.040> of<00:13:48.190> FPGAs 00:13:48.870 --> 00:13:48.880 align:start position:0% actually what the acorn line of FPGAs 00:13:48.880 --> 00:13:51.590 align:start position:0% actually what the acorn line of FPGAs are<00:13:49.060> designed<00:13:49.390> to<00:13:49.630> do<00:13:49.779> offload<00:13:50.740> catch<00:13:51.339> hack 00:13:51.590 --> 00:13:51.600 align:start position:0% are designed to do offload catch hack 00:13:51.600 --> 00:13:53.940 align:start position:0% are designed to do offload catch hack computations<00:13:52.600> to<00:13:52.720> save<00:13:53.020> system<00:13:53.440> power<00:13:53.650> and 00:13:53.940 --> 00:13:53.950 align:start position:0% computations to save system power and 00:13:53.950 --> 00:13:57.060 align:start position:0% computations to save system power and increase<00:13:54.760> performance<00:13:55.650> you<00:13:56.650> shouldn't<00:13:56.950> be 00:13:57.060 --> 00:13:57.070 align:start position:0% increase performance you shouldn't be 00:13:57.070 --> 00:13:59.100 align:start position:0% increase performance you shouldn't be forced<00:13:57.490> into<00:13:57.670> buying<00:13:58.000> extensions<00:13:58.779> to<00:13:58.960> your 00:13:59.100 --> 00:13:59.110 align:start position:0% forced into buying extensions to your 00:13:59.110 --> 00:14:01.079 align:start position:0% forced into buying extensions to your hardware<00:13:59.440> just<00:14:00.250> to<00:14:00.400> compete<00:14:00.760> with<00:14:00.880> other 00:14:01.079 --> 00:14:01.089 align:start position:0% hardware just to compete with other 00:14:01.089 --> 00:14:04.950 align:start position:0% hardware just to compete with other hardware<00:14:02.220> but<00:14:03.220> hey<00:14:03.370> guess<00:14:03.850> what<00:14:04.060> Asics<00:14:04.660> can<00:14:04.839> do 00:14:04.950 --> 00:14:04.960 align:start position:0% hardware but hey guess what Asics can do 00:14:04.960 --> 00:14:07.860 align:start position:0% hardware but hey guess what Asics can do that<00:14:05.110> too<00:14:05.940> profiling<00:14:06.940> at<00:14:07.089> hash<00:14:07.300> with<00:14:07.600> catch 00:14:07.860 --> 00:14:07.870 align:start position:0% that too profiling at hash with catch 00:14:07.870 --> 00:14:10.050 align:start position:0% that too profiling at hash with catch Accra<00:14:08.230> moved<00:14:08.470> shows<00:14:09.190> us<00:14:09.430> that<00:14:09.580> the<00:14:09.700> compute 00:14:10.050 --> 00:14:10.060 align:start position:0% Accra moved shows us that the compute 00:14:10.060 --> 00:14:11.880 align:start position:0% Accra moved shows us that the compute cause<00:14:10.270> of<00:14:10.480> a<00:14:10.540> car<00:14:10.570> to<00:14:10.960> really<00:14:11.260> only<00:14:11.500> utilize 00:14:11.880 --> 00:14:11.890 align:start position:0% cause of a car to really only utilize 00:14:11.890 --> 00:14:14.610 align:start position:0% cause of a car to really only utilize 20%<00:14:12.580> of<00:14:12.670> the<00:14:12.730> time<00:14:12.940> allowing<00:14:13.900> for<00:14:14.230> 10% 00:14:14.610 --> 00:14:14.620 align:start position:0% 20% of the time allowing for 10% 00:14:14.620 --> 00:14:17.250 align:start position:0% 20% of the time allowing for 10% efficiency<00:14:15.310> gain<00:14:15.490> right<00:14:15.730> off<00:14:15.850> the<00:14:15.970> bat<00:14:16.260> simile 00:14:17.250 --> 00:14:17.260 align:start position:0% efficiency gain right off the bat simile 00:14:17.260 --> 00:14:20.700 align:start position:0% efficiency gain right off the bat simile our<00:14:17.500> personal<00:14:18.070> /<00:14:18.790> sauce<00:14:19.089> line<00:14:19.750> profiling<00:14:20.589> up 00:14:20.700 --> 00:14:20.710 align:start position:0% our personal / sauce line profiling up 00:14:20.710 --> 00:14:23.880 align:start position:0% our personal / sauce line profiling up there<00:14:21.180> shows<00:14:22.180> us<00:14:22.390> that<00:14:22.540> more<00:14:22.779> than<00:14:22.959> 20%<00:14:23.709> of<00:14:23.770> the 00:14:23.880 --> 00:14:23.890 align:start position:0% there shows us that more than 20% of the 00:14:23.890 --> 00:14:26.160 align:start position:0% there shows us that more than 20% of the all<00:14:24.010> instructions<00:14:24.970> are<00:14:25.120> catch<00:14:25.330> shock<00:14:25.540> which 00:14:26.160 --> 00:14:26.170 align:start position:0% all instructions are catch shock which 00:14:26.170 --> 00:14:29.340 align:start position:0% all instructions are catch shock which can<00:14:26.410> all<00:14:26.500> be<00:14:26.529> offloaded<00:14:27.240> that's<00:14:28.240> a<00:14:28.390> 30% 00:14:29.340 --> 00:14:29.350 align:start position:0% can all be offloaded that's a 30% 00:14:29.350 --> 00:14:35.579 align:start position:0% can all be offloaded that's a 30% performance<00:14:29.920> gain<00:14:30.130> folks<00:14:30.870> 30%<00:14:33.270> okay<00:14:34.300> so<00:14:35.290> I've 00:14:35.579 --> 00:14:35.589 align:start position:0% performance gain folks 30% okay so I've 00:14:35.589 --> 00:14:36.660 align:start position:0% performance gain folks 30% okay so I've talked<00:14:35.830> a<00:14:35.920> little<00:14:36.070> bit<00:14:36.310> about<00:14:36.459> the 00:14:36.660 --> 00:14:36.670 align:start position:0% talked a little bit about the 00:14:36.670 --> 00:14:38.340 align:start position:0% talked a little bit about the deficiencies<00:14:37.240> about<00:14:37.480> hash<00:14:37.660> so<00:14:37.900> what<00:14:38.080> the<00:14:38.200> hell 00:14:38.340 --> 00:14:38.350 align:start position:0% deficiencies about hash so what the hell 00:14:38.350 --> 00:14:39.990 align:start position:0% deficiencies about hash so what the hell did<00:14:38.500> we<00:14:38.560> do<00:14:38.770> about<00:14:39.010> it<00:14:39.190> we're<00:14:39.520> just<00:14:39.550> gonna<00:14:39.730> sit 00:14:39.990 --> 00:14:40.000 align:start position:0% did we do about it we're just gonna sit 00:14:40.000 --> 00:14:43.050 align:start position:0% did we do about it we're just gonna sit up<00:14:40.060> here<00:14:40.240> in<00:14:40.300> trash<00:14:40.570> at<00:14:40.750> hash<00:14:40.959> no<00:14:41.640> we<00:14:42.640> went<00:14:42.850> and 00:14:43.050 --> 00:14:43.060 align:start position:0% up here in trash at hash no we went and 00:14:43.060 --> 00:14:47.190 align:start position:0% up here in trash at hash no we went and kind<00:14:43.570> of<00:14:43.690> fixed<00:14:43.930> it<00:14:44.050> so<00:14:45.330> that's<00:14:46.330> a<00:14:46.480> pro<00:14:46.839> growl 00:14:47.190 --> 00:14:47.200 align:start position:0% kind of fixed it so that's a pro growl 00:14:47.200 --> 00:14:49.829 align:start position:0% kind of fixed it so that's a pro growl spaghetti<00:14:47.800> and<00:14:47.950> Lego<00:14:48.279> this<00:14:49.209> was<00:14:49.420> made<00:14:49.630> on<00:14:49.810> a 00:14:49.829 --> 00:14:49.839 align:start position:0% spaghetti and Lego this was made on a 00:14:49.839 --> 00:14:51.780 align:start position:0% spaghetti and Lego this was made on a plane<00:14:50.200> while<00:14:50.470> I<00:14:50.529> was<00:14:50.770> slightly<00:14:51.040> drunk<00:14:51.490> so 00:14:51.780 --> 00:14:51.790 align:start position:0% plane while I was slightly drunk so 00:14:51.790 --> 00:14:54.840 align:start position:0% plane while I was slightly drunk so excuse<00:14:52.270> the<00:14:52.300> mess<00:14:53.130> there<00:14:54.130> are<00:14:54.250> five<00:14:54.550> major 00:14:54.840 --> 00:14:54.850 align:start position:0% excuse the mess there are five major 00:14:54.850 --> 00:14:57.930 align:start position:0% excuse the mess there are five major changes<00:14:55.420> here<00:14:55.660> see<00:14:56.650> we<00:14:56.770> changed<00:14:57.100> cat<00:14:57.700> jack 00:14:57.930 --> 00:14:57.940 align:start position:0% changes here see we changed cat jack 00:14:57.940 --> 00:15:01.560 align:start position:0% changes here see we changed cat jack from<00:14:58.270> F<00:14:58.510> 1600<00:14:59.440> to<00:14:59.560> f800<00:15:00.279> I'll<00:15:01.000> go<00:15:01.209> into<00:15:01.420> why 00:15:01.560 --> 00:15:01.570 align:start position:0% from F 1600 to f800 I'll go into why 00:15:01.570 --> 00:15:03.870 align:start position:0% from F 1600 to f800 I'll go into why that<00:15:01.600> was<00:15:01.930> really<00:15:02.200> important<00:15:02.620> in<00:15:02.680> a<00:15:02.740> moment<00:15:02.920> we 00:15:03.870 --> 00:15:03.880 align:start position:0% that was really important in a moment we 00:15:03.880 --> 00:15:05.820 align:start position:0% that was really important in a moment we increased<00:15:04.240> the<00:15:04.390> mix<00:15:04.600> state<00:15:04.900> we<00:15:05.470> added<00:15:05.740> a 00:15:05.820 --> 00:15:05.830 align:start position:0% increased the mix state we added a 00:15:05.830 --> 00:15:07.790 align:start position:0% increased the mix state we added a random<00:15:06.220> sequence<00:15:06.430> of<00:15:06.760> math<00:15:06.940> in<00:15:07.240> the<00:15:07.330> main<00:15:07.510> loop 00:15:07.790 --> 00:15:07.800 align:start position:0% random sequence of math in the main loop 00:15:07.800 --> 00:15:10.890 align:start position:0% random sequence of math in the main loop we<00:15:08.800> added<00:15:09.070> reads<00:15:09.370> from<00:15:09.670> a<00:15:09.820> small<00:15:10.089> low<00:15:10.300> latency 00:15:10.890 --> 00:15:10.900 align:start position:0% we added reads from a small low latency 00:15:10.900 --> 00:15:13.710 align:start position:0% we added reads from a small low latency cache<00:15:11.709> that<00:15:12.010> supports<00:15:12.430> random<00:15:12.790> addresses<00:15:12.970> and 00:15:13.710 --> 00:15:13.720 align:start position:0% cache that supports random addresses and 00:15:13.720 --> 00:15:16.860 align:start position:0% cache that supports random addresses and we<00:15:14.200> increased<00:15:14.560> the<00:15:14.709> DRM<00:15:15.100> read<00:15:15.670> from<00:15:16.060> 128<00:15:16.510> bytes 00:15:16.860 --> 00:15:16.870 align:start position:0% we increased the DRM read from 128 bytes 00:15:16.870 --> 00:15:21.210 align:start position:0% we increased the DRM read from 128 bytes to<00:15:17.620> 256<00:15:18.310> bytes<00:15:18.550> and<00:15:19.709> this<00:15:20.709> is<00:15:20.860> what<00:15:21.010> it<00:15:21.100> all 00:15:21.210 --> 00:15:21.220 align:start position:0% to 256 bytes and this is what it all 00:15:21.220 --> 00:15:26.430 align:start position:0% to 256 bytes and this is what it all became<00:15:23.430> see<00:15:24.430> the<00:15:25.120> most<00:15:25.420> efficient<00:15:26.290> and 00:15:26.430 --> 00:15:26.440 align:start position:0% became see the most efficient and 00:15:26.440 --> 00:15:28.710 align:start position:0% became see the most efficient and effective<00:15:26.740> algorithms<00:15:27.700> are<00:15:28.240> the<00:15:28.450> simplest 00:15:28.710 --> 00:15:28.720 align:start position:0% effective algorithms are the simplest 00:15:28.720 --> 00:15:31.769 align:start position:0% effective algorithms are the simplest and<00:15:29.500> any<00:15:29.800> ASIC<00:15:30.250> designer<00:15:30.579> will<00:15:30.760> tell<00:15:30.910> you<00:15:31.030> that 00:15:31.769 --> 00:15:31.779 align:start position:0% and any ASIC designer will tell you that 00:15:31.779 --> 00:15:33.629 align:start position:0% and any ASIC designer will tell you that complexity<00:15:32.230> leads<00:15:32.649> to<00:15:32.920> weakness<00:15:33.100> in 00:15:33.629 --> 00:15:33.639 align:start position:0% complexity leads to weakness in 00:15:33.639 --> 00:15:37.679 align:start position:0% complexity leads to weakness in proof-of-work<00:15:35.160> okay<00:15:36.160> so<00:15:36.790> why<00:15:36.939> these<00:15:37.149> changes 00:15:37.679 --> 00:15:37.689 align:start position:0% proof-of-work okay so why these changes 00:15:37.689 --> 00:15:40.829 align:start position:0% proof-of-work okay so why these changes well<00:15:38.680> let's<00:15:38.980> stop<00:15:39.220> with<00:15:39.639> cat<00:15:40.120> check<00:15:40.360> and<00:15:40.600> dear 00:15:40.829 --> 00:15:40.839 align:start position:0% well let's stop with cat check and dear 00:15:40.839 --> 00:15:44.040 align:start position:0% well let's stop with cat check and dear M<00:15:40.990> read<00:15:41.819> catch<00:15:42.819> hack<00:15:42.999> hashes<00:15:43.480> used<00:15:43.839> at<00:15:43.959> the 00:15:44.040 --> 00:15:44.050 align:start position:0% M read catch hack hashes used at the 00:15:44.050 --> 00:15:45.900 align:start position:0% M read catch hack hashes used at the start<00:15:44.259> and<00:15:44.470> end<00:15:44.649> of<00:15:44.740> it<00:15:44.860> hash<00:15:45.100> get<00:15:45.550> reduced 00:15:45.900 --> 00:15:45.910 align:start position:0% start and end of it hash get reduced 00:15:45.910 --> 00:15:50.220 align:start position:0% start and end of it hash get reduced from<00:15:46.120> F<00:15:46.329> 1600<00:15:47.170> with<00:15:47.560> a<00:15:47.589> ward<00:15:47.889> size<00:15:48.129> of<00:15:48.339> 64<00:15:48.730> f<00:15:49.569> 800 00:15:50.220 --> 00:15:50.230 align:start position:0% from F 1600 with a ward size of 64 f 800 00:15:50.230 --> 00:15:52.590 align:start position:0% from F 1600 with a ward size of 64 f 800 with<00:15:50.379> a<00:15:50.410> word<00:15:50.620> size<00:15:50.889> of<00:15:51.069> 32<00:15:51.370> why 00:15:52.590 --> 00:15:52.600 align:start position:0% with a word size of 32 why 00:15:52.600 --> 00:15:55.799 align:start position:0% with a word size of 32 why well<00:15:53.350> GPUs<00:15:54.069> actually<00:15:54.430> have<00:15:54.550> 32-bit<00:15:55.269> data<00:15:55.300> pods 00:15:55.799 --> 00:15:55.809 align:start position:0% well GPUs actually have 32-bit data pods 00:15:55.809 --> 00:15:59.220 align:start position:0% well GPUs actually have 32-bit data pods and<00:15:56.199> F<00:15:57.129> 1600<00:15:58.120> requires<00:15:58.480> twice<00:15:58.839> as<00:15:59.079> many 00:15:59.220 --> 00:15:59.230 align:start position:0% and F 1600 requires twice as many 00:15:59.230 --> 00:16:01.019 align:start position:0% and F 1600 requires twice as many instructions<00:15:59.949> to<00:16:00.069> execute<00:16:00.550> on<00:16:00.639> a<00:16:00.699> graphic 00:16:01.019 --> 00:16:01.029 align:start position:0% instructions to execute on a graphic 00:16:01.029 --> 00:16:04.470 align:start position:0% instructions to execute on a graphic card<00:16:01.290> it's<00:16:02.290> it's<00:16:02.649> wasted<00:16:03.160> cycles<00:16:03.370> yet<00:16:04.209> hash 00:16:04.470 --> 00:16:04.480 align:start position:0% card it's it's wasted cycles yet hash 00:16:04.480 --> 00:16:07.019 align:start position:0% card it's it's wasted cycles yet hash does<00:16:04.749> not<00:16:05.019> use<00:16:05.410> the<00:16:05.620> extra<00:16:06.040> data<00:16:06.249> processed<00:16:06.850> by 00:16:07.019 --> 00:16:07.029 align:start position:0% does not use the extra data processed by 00:16:07.029 --> 00:16:10.139 align:start position:0% does not use the extra data processed by F<00:16:07.180> 1600<00:16:07.990> so<00:16:08.499> reducing<00:16:09.399> this<00:16:09.519> amount<00:16:09.819> of<00:16:09.910> data 00:16:10.139 --> 00:16:10.149 align:start position:0% F 1600 so reducing this amount of data 00:16:10.149 --> 00:16:11.999 align:start position:0% F 1600 so reducing this amount of data has<00:16:10.480> no<00:16:10.779> effect<00:16:11.170> on<00:16:11.290> the<00:16:11.410> security<00:16:11.829> of<00:16:11.860> the 00:16:11.999 --> 00:16:12.009 align:start position:0% has no effect on the security of the 00:16:12.009 --> 00:16:15.660 align:start position:0% has no effect on the security of the algorithm<00:16:12.540> but<00:16:13.540> what<00:16:13.720> it<00:16:13.839> does<00:16:14.050> do<00:16:14.350> is<00:16:14.670> reduces 00:16:15.660 --> 00:16:15.670 align:start position:0% algorithm but what it does do is reduces 00:16:15.670 --> 00:16:18.090 align:start position:0% algorithm but what it does do is reduces any<00:16:15.939> possible<00:16:16.899> efficiency<00:16:17.499> gains<00:16:17.740> from 00:16:18.090 --> 00:16:18.100 align:start position:0% any possible efficiency gains from 00:16:18.100 --> 00:16:20.129 align:start position:0% any possible efficiency gains from offloading<00:16:18.819> the<00:16:19.029> kept<00:16:19.209> track<00:16:19.420> computations 00:16:20.129 --> 00:16:20.139 align:start position:0% offloading the kept track computations 00:16:20.139 --> 00:16:24.239 align:start position:0% offloading the kept track computations from<00:16:20.920> the<00:16:21.040> GPU<00:16:22.139> that<00:16:23.139> eliminates<00:16:23.649> your<00:16:23.829> a<00:16:23.889> 6 00:16:24.239 --> 00:16:24.249 align:start position:0% from the GPU that eliminates your a 6 00:16:24.249 --> 00:16:27.540 align:start position:0% from the GPU that eliminates your a 6 speed<00:16:24.370> up<00:16:24.610> that<00:16:25.120> eliminates<00:16:25.629> your<00:16:26.069> your<00:16:27.069> FPGA 00:16:27.540 --> 00:16:27.550 align:start position:0% speed up that eliminates your your FPGA 00:16:27.550 --> 00:16:29.040 align:start position:0% speed up that eliminates your your FPGA speed<00:16:27.910> up<00:16:28.029> and<00:16:28.269> I<00:16:28.360> should<00:16:28.509> have<00:16:28.569> just<00:16:28.749> put<00:16:28.930> that 00:16:29.040 --> 00:16:29.050 align:start position:0% speed up and I should have just put that 00:16:29.050 --> 00:16:29.400 align:start position:0% speed up and I should have just put that out<00:16:29.110> there 00:16:29.400 --> 00:16:29.410 align:start position:0% out there 00:16:29.410 --> 00:16:36.929 align:start position:0% out there oops<00:16:29.759> so<00:16:30.759> by<00:16:30.939> by<00:16:31.120> 30%<00:16:32.129> performance<00:16:33.129> again<00:16:35.939> we 00:16:36.929 --> 00:16:36.939 align:start position:0% oops so by by 30% performance again we 00:16:36.939 --> 00:16:38.850 align:start position:0% oops so by by 30% performance again we leave<00:16:37.180> the<00:16:37.329> number<00:16:37.600> of<00:16:37.689> accesses<00:16:38.199> to<00:16:38.379> the<00:16:38.589> dag 00:16:38.850 --> 00:16:38.860 align:start position:0% leave the number of accesses to the dag 00:16:38.860 --> 00:16:40.350 align:start position:0% leave the number of accesses to the dag which<00:16:39.279> is<00:16:39.430> also<00:16:39.579> the<00:16:39.730> number<00:16:40.029> of<00:16:40.059> the<00:16:40.149> loop 00:16:40.350 --> 00:16:40.360 align:start position:0% which is also the number of the loop 00:16:40.360 --> 00:16:42.869 align:start position:0% which is also the number of the loop iterations<00:16:40.930> unchanged<00:16:41.680> from<00:16:41.740> at<00:16:41.949> hashes<00:16:42.309> 64 00:16:42.869 --> 00:16:42.879 align:start position:0% iterations unchanged from at hashes 64 00:16:42.879 --> 00:16:44.579 align:start position:0% iterations unchanged from at hashes 64 we<00:16:43.269> don't<00:16:43.449> need<00:16:43.540> to<00:16:43.689> touch<00:16:43.870> that<00:16:44.110> don't<00:16:44.500> need 00:16:44.579 --> 00:16:44.589 align:start position:0% we don't need to touch that don't need 00:16:44.589 --> 00:16:46.679 align:start position:0% we don't need to touch that don't need to<00:16:44.709> dick<00:16:44.829> around<00:16:44.860> with<00:16:45.189> it<00:16:45.399> but<00:16:46.120> the<00:16:46.209> dag<00:16:46.389> reads 00:16:46.679 --> 00:16:46.689 align:start position:0% to dick around with it but the dag reads 00:16:46.689 --> 00:16:50.490 align:start position:0% to dick around with it but the dag reads size<00:16:47.100> we<00:16:48.100> increase<00:16:48.459> from<00:16:48.639> 128<00:16:49.269> bytes<00:16:49.509> to<00:16:49.839> 256 00:16:50.490 --> 00:16:50.500 align:start position:0% size we increase from 128 bytes to 256 00:16:50.500 --> 00:16:52.920 align:start position:0% size we increase from 128 bytes to 256 bytes<00:16:50.740> this<00:16:51.699> allows<00:16:52.000> prog<00:16:52.360> cow<00:16:52.569> to<00:16:52.899> be 00:16:52.920 --> 00:16:52.930 align:start position:0% bytes this allows prog cow to be 00:16:52.930 --> 00:16:55.199 align:start position:0% bytes this allows prog cow to be efficiently<00:16:53.439> executed<00:16:54.129> on<00:16:54.250> all<00:16:54.550> current<00:16:55.120> and 00:16:55.199 --> 00:16:55.209 align:start position:0% efficiently executed on all current and 00:16:55.209 --> 00:16:57.569 align:start position:0% efficiently executed on all current and hopefully<00:16:55.809> near<00:16:56.079> future<00:16:56.410> DRAM<00:16:56.949> technologies 00:16:57.569 --> 00:16:57.579 align:start position:0% hopefully near future DRAM technologies 00:16:57.579 --> 00:17:00.840 align:start position:0% hopefully near future DRAM technologies without<00:16:58.120> requiring<00:16:58.720> overclocking<00:16:59.850> that 00:17:00.840 --> 00:17:00.850 align:start position:0% without requiring overclocking that 00:17:00.850 --> 00:17:02.819 align:start position:0% without requiring overclocking that means<00:17:01.089> there's<00:17:01.420> no<00:17:01.600> more<00:17:01.629> no<00:17:02.290> more<00:17:02.470> system 00:17:02.819 --> 00:17:02.829 align:start position:0% means there's no more no more system 00:17:02.829 --> 00:17:04.980 align:start position:0% means there's no more no more system tuning<00:17:03.220> it<00:17:03.699> just<00:17:03.939> works<00:17:04.120> off<00:17:04.329> the<00:17:04.480> bat<00:17:04.630> fully 00:17:04.980 --> 00:17:04.990 align:start position:0% tuning it just works off the bat fully 00:17:04.990 --> 00:17:12.799 align:start position:0% tuning it just works off the bat fully tuned<00:17:05.319> fully<00:17:05.620> optimized<00:17:10.740> now<00:17:11.740> for<00:17:12.189> this<00:17:12.309> part 00:17:12.799 --> 00:17:12.809 align:start position:0% tuned fully optimized now for this part 00:17:12.809 --> 00:17:16.319 align:start position:0% tuned fully optimized now for this part see<00:17:13.809> as<00:17:14.110> I<00:17:14.439> said<00:17:14.500> GPU<00:17:15.370> cores<00:17:15.669> are<00:17:15.970> most 00:17:16.319 --> 00:17:16.329 align:start position:0% see as I said GPU cores are most 00:17:16.329 --> 00:17:18.689 align:start position:0% see as I said GPU cores are most efficient<00:17:16.750> when<00:17:17.020> they're<00:17:17.199> doing<00:17:17.500> 16<00:17:18.280> byte<00:17:18.490> or 00:17:18.689 --> 00:17:18.699 align:start position:0% efficient when they're doing 16 byte or 00:17:18.699 --> 00:17:22.130 align:start position:0% efficient when they're doing 16 byte or forward<00:17:19.270> loads<00:17:19.569> and<00:17:19.990> in<00:17:20.949> order<00:17:21.220> to<00:17:21.250> have<00:17:21.520> our 00:17:22.130 --> 00:17:22.140 align:start position:0% forward loads and in order to have our 00:17:22.140 --> 00:17:26.000 align:start position:0% forward loads and in order to have our 256<00:17:23.140> byte<00:17:23.319> loads<00:17:23.589> we<00:17:23.829> have<00:17:23.980> to<00:17:24.100> do<00:17:24.480> 256<00:17:25.480> bytes 00:17:26.000 --> 00:17:26.010 align:start position:0% 256 byte loads we have to do 256 bytes 00:17:26.010 --> 00:17:29.220 align:start position:0% 256 byte loads we have to do 256 bytes by<00:17:27.010> 16<00:17:27.429> bytes<00:17:27.640> per<00:17:27.850> Lane<00:17:28.059> which<00:17:28.720> ends<00:17:28.990> up<00:17:29.080> with 00:17:29.220 --> 00:17:29.230 align:start position:0% by 16 bytes per Lane which ends up with 00:17:29.230 --> 00:17:32.720 align:start position:0% by 16 bytes per Lane which ends up with 16<00:17:29.649> lanes<00:17:29.860> working<00:17:30.370> together<00:17:30.580> in<00:17:30.850> parallel 00:17:32.720 --> 00:17:32.730 align:start position:0% 16 lanes working together in parallel 00:17:32.730 --> 00:17:35.580 align:start position:0% 16 lanes working together in parallel yeah<00:17:33.760> if<00:17:34.030> I'm<00:17:34.330> getting<00:17:34.450> too<00:17:34.720> technical<00:17:34.899> let<00:17:35.470> me 00:17:35.580 --> 00:17:35.590 align:start position:0% yeah if I'm getting too technical let me 00:17:35.590 --> 00:17:37.830 align:start position:0% yeah if I'm getting too technical let me know<00:17:35.820> when<00:17:36.820> we<00:17:36.909> work<00:17:37.090> back<00:17:37.330> from<00:17:37.510> the<00:17:37.630> frame 00:17:37.830 --> 00:17:37.840 align:start position:0% know when we work back from the frame 00:17:37.840 --> 00:17:40.919 align:start position:0% know when we work back from the frame buffer<00:17:38.049> interface<00:17:38.590> GPUs<00:17:39.490> have<00:17:39.520> an<00:17:40.090> l2<00:17:40.480> cache 00:17:40.919 --> 00:17:40.929 align:start position:0% buffer interface GPUs have an l2 cache 00:17:40.929 --> 00:17:44.279 align:start position:0% buffer interface GPUs have an l2 cache an<00:17:41.289> l1<00:17:41.710> cache<00:17:41.980> and<00:17:42.340> a<00:17:42.789> texture<00:17:43.179> cache<00:17:43.360> we<00:17:44.110> use 00:17:44.279 --> 00:17:44.289 align:start position:0% an l1 cache and a texture cache we use 00:17:44.289 --> 00:17:45.300 align:start position:0% an l1 cache and a texture cache we use these<00:17:44.500> things<00:17:44.559> for 00:17:45.300 --> 00:17:45.310 align:start position:0% these things for 00:17:45.310 --> 00:17:48.870 align:start position:0% these things for graphics<00:17:45.760> and<00:17:46.030> virtualization<00:17:47.400> we<00:17:48.400> haven't 00:17:48.870 --> 00:17:48.880 align:start position:0% graphics and virtualization we haven't 00:17:48.880 --> 00:17:51.390 align:start position:0% graphics and virtualization we haven't discovered<00:17:49.300> a<00:17:49.540> way<00:17:49.750> or<00:17:50.110> a<00:17:50.170> method<00:17:51.010> of<00:17:51.100> making 00:17:51.390 --> 00:17:51.400 align:start position:0% discovered a way or a method of making 00:17:51.400 --> 00:17:53.880 align:start position:0% discovered a way or a method of making use<00:17:51.790> of<00:17:51.820> these<00:17:52.090> cases<00:17:52.420> that's<00:17:53.110> both<00:17:53.380> efficient 00:17:53.880 --> 00:17:53.890 align:start position:0% use of these cases that's both efficient 00:17:53.890 --> 00:17:56.640 align:start position:0% use of these cases that's both efficient and<00:17:54.040> portable<00:17:54.490> across<00:17:54.700> GPU<00:17:55.330> architectures<00:17:56.020> so 00:17:56.640 --> 00:17:56.650 align:start position:0% and portable across GPU architectures so 00:17:56.650 --> 00:17:58.980 align:start position:0% and portable across GPU architectures so we<00:17:56.770> leave<00:17:56.980> those<00:17:57.160> alone<00:17:57.390> that<00:17:58.390> does<00:17:58.630> allow<00:17:58.960> for 00:17:58.980 --> 00:17:58.990 align:start position:0% we leave those alone that does allow for 00:17:58.990 --> 00:18:01.320 align:start position:0% we leave those alone that does allow for a<00:17:59.320> potential<00:18:00.010> efficiency<00:18:00.490> gain<00:18:00.670> in<00:18:00.910> a<00:18:01.030> fixed 00:18:01.320 --> 00:18:01.330 align:start position:0% a potential efficiency gain in a fixed 00:18:01.330 --> 00:18:02.670 align:start position:0% a potential efficiency gain in a fixed function<00:18:01.480> ASIC<00:18:02.050> where<00:18:02.200> you<00:18:02.290> could<00:18:02.440> shave 00:18:02.670 --> 00:18:02.680 align:start position:0% function ASIC where you could shave 00:18:02.680 --> 00:18:04.440 align:start position:0% function ASIC where you could shave those<00:18:02.890> off<00:18:03.100> and<00:18:03.370> we'll<00:18:03.760> touch<00:18:03.910> in<00:18:04.120> that<00:18:04.240> on<00:18:04.360> the 00:18:04.440 --> 00:18:04.450 align:start position:0% those off and we'll touch in that on the 00:18:04.450 --> 00:18:07.380 align:start position:0% those off and we'll touch in that on the breakout<00:18:04.810> session<00:18:05.430> so<00:18:06.430> propel<00:18:06.940> doesn't 00:18:07.380 --> 00:18:07.390 align:start position:0% breakout session so propel doesn't 00:18:07.390 --> 00:18:09.600 align:start position:0% breakout session so propel doesn't target<00:18:07.570> those<00:18:08.290> cases<00:18:08.830> it<00:18:09.040> simply<00:18:09.310> passes 00:18:09.600 --> 00:18:09.610 align:start position:0% target those cases it simply passes 00:18:09.610 --> 00:18:12.450 align:start position:0% target those cases it simply passes through<00:18:10.090> the<00:18:10.210> dag<00:18:10.390> load<00:18:10.660> but<00:18:11.530> next<00:18:12.100> to<00:18:12.370> the 00:18:12.450 --> 00:18:12.460 align:start position:0% through the dag load but next to the 00:18:12.460 --> 00:18:14.850 align:start position:0% through the dag load but next to the cases<00:18:12.910> we<00:18:13.450> have<00:18:13.630> something<00:18:14.110> that<00:18:14.260> I<00:18:14.590> love 00:18:14.850 --> 00:18:14.860 align:start position:0% cases we have something that I love 00:18:14.860 --> 00:18:19.080 align:start position:0% cases we have something that I love which<00:18:15.610> is<00:18:15.640> called<00:18:16.650> scratchpad<00:18:17.650> memory<00:18:17.940> so<00:18:18.940> let 00:18:19.080 --> 00:18:19.090 align:start position:0% which is called scratchpad memory so let 00:18:19.090 --> 00:18:19.950 align:start position:0% which is called scratchpad memory so let me<00:18:19.180> tell<00:18:19.300> you<00:18:19.420> a<00:18:19.480> little<00:18:19.750> bit<00:18:19.840> about 00:18:19.950 --> 00:18:19.960 align:start position:0% me tell you a little bit about 00:18:19.960 --> 00:18:22.130 align:start position:0% me tell you a little bit about scratchpad<00:18:20.380> memory<00:18:20.620> it's<00:18:21.490> this<00:18:21.730> high<00:18:21.970> speed 00:18:22.130 --> 00:18:22.140 align:start position:0% scratchpad memory it's this high speed 00:18:22.140 --> 00:18:24.420 align:start position:0% scratchpad memory it's this high speed internal<00:18:23.140> memory<00:18:23.530> used<00:18:23.740> for<00:18:23.980> temporary 00:18:24.420 --> 00:18:24.430 align:start position:0% internal memory used for temporary 00:18:24.430 --> 00:18:27.180 align:start position:0% internal memory used for temporary storage<00:18:24.850> of<00:18:25.240> calculations<00:18:25.900> data<00:18:26.230> other<00:18:26.860> works 00:18:27.180 --> 00:18:27.190 align:start position:0% storage of calculations data other works 00:18:27.190 --> 00:18:27.840 align:start position:0% storage of calculations data other works in<00:18:27.400> progress 00:18:27.840 --> 00:18:27.850 align:start position:0% in progress 00:18:27.850 --> 00:18:30.570 align:start position:0% in progress think<00:18:28.330> of<00:18:28.420> it<00:18:28.510> like<00:18:28.600> your<00:18:28.780> brain<00:18:29.430> Nvidia<00:18:30.430> and 00:18:30.570 --> 00:18:30.580 align:start position:0% think of it like your brain Nvidia and 00:18:30.580 --> 00:18:32.580 align:start position:0% think of it like your brain Nvidia and q2<00:18:30.850> referred<00:18:31.150> to<00:18:31.270> this<00:18:31.360> as<00:18:31.570> shared<00:18:31.840> memory<00:18:32.290> and 00:18:32.580 --> 00:18:32.590 align:start position:0% q2 referred to this as shared memory and 00:18:32.590 --> 00:18:35.400 align:start position:0% q2 referred to this as shared memory and AMD<00:18:33.490> and<00:18:33.730> OpenCL<00:18:34.150> refer<00:18:34.630> to<00:18:34.780> this<00:18:34.900> as<00:18:35.080> local 00:18:35.400 --> 00:18:35.410 align:start position:0% AMD and OpenCL refer to this as local 00:18:35.410 --> 00:18:37.770 align:start position:0% AMD and OpenCL refer to this as local memory<00:18:35.620> the<00:18:36.580> defining<00:18:37.090> feature<00:18:37.360> of<00:18:37.540> this 00:18:37.770 --> 00:18:37.780 align:start position:0% memory the defining feature of this 00:18:37.780 --> 00:18:40.110 align:start position:0% memory the defining feature of this memory<00:18:38.050> compared<00:18:38.770> to<00:18:38.890> DRAM<00:18:39.280> is<00:18:39.460> that<00:18:39.970> it's 00:18:40.110 --> 00:18:40.120 align:start position:0% memory compared to DRAM is that it's 00:18:40.120 --> 00:18:42.510 align:start position:0% memory compared to DRAM is that it's highly<00:18:40.540> banked<00:18:40.840> with<00:18:40.930> a<00:18:40.990> large<00:18:41.260> crossbar<00:18:41.830> that 00:18:42.510 --> 00:18:42.520 align:start position:0% highly banked with a large crossbar that 00:18:42.520 --> 00:18:44.550 align:start position:0% highly banked with a large crossbar that means<00:18:42.730> it<00:18:42.880> allows<00:18:42.970> accesses<00:18:43.870> to<00:18:44.080> random 00:18:44.550 --> 00:18:44.560 align:start position:0% means it allows accesses to random 00:18:44.560 --> 00:18:46.860 align:start position:0% means it allows accesses to random addresses<00:18:44.800> to<00:18:45.250> be<00:18:45.370> processed<00:18:45.880> really<00:18:46.390> quickly 00:18:46.860 --> 00:18:46.870 align:start position:0% addresses to be processed really quickly 00:18:46.870 --> 00:18:48.750 align:start position:0% addresses to be processed really quickly something<00:18:47.530> that<00:18:47.650> fixed<00:18:48.010> function<00:18:48.190> Hardware 00:18:48.750 --> 00:18:48.760 align:start position:0% something that fixed function Hardware 00:18:48.760 --> 00:18:51.060 align:start position:0% something that fixed function Hardware it<00:18:48.850> doesn't<00:18:49.120> do<00:18:49.210> too<00:18:49.360> well<00:18:49.390> with<00:18:50.010> Nvidia<00:18:51.010> is 00:18:51.060 --> 00:18:51.070 align:start position:0% it doesn't do too well with Nvidia is 00:18:51.070 --> 00:18:53.520 align:start position:0% it doesn't do too well with Nvidia is Pascal<00:18:51.580> outline<00:18:52.150> supports<00:18:52.810> a<00:18:52.930> scratch<00:18:53.170> pad<00:18:53.380> of 00:18:53.520 --> 00:18:53.530 align:start position:0% Pascal outline supports a scratch pad of 00:18:53.530 --> 00:18:56.220 align:start position:0% Pascal outline supports a scratch pad of up<00:18:53.650> to<00:18:53.680> 96<00:18:54.310> kilobytes<00:18:54.760> AMD's<00:18:55.600> Polaris<00:18:56.080> and 00:18:56.220 --> 00:18:56.230 align:start position:0% up to 96 kilobytes AMD's Polaris and 00:18:56.230 --> 00:18:58.800 align:start position:0% up to 96 kilobytes AMD's Polaris and Vega<00:18:56.440> line<00:18:56.710> supports<00:18:57.610> up<00:18:57.730> to<00:18:57.760> 64<00:18:58.330> kilobytes 00:18:58.800 --> 00:18:58.810 align:start position:0% Vega line supports up to 64 kilobytes 00:18:58.810 --> 00:19:01.980 align:start position:0% Vega line supports up to 64 kilobytes and<00:18:59.110> the<00:18:59.980> AMD<00:19:00.430> OpenCL<00:19:01.150> kernel<00:19:01.360> currently 00:19:01.980 --> 00:19:01.990 align:start position:0% and the AMD OpenCL kernel currently 00:19:01.990 --> 00:19:04.080 align:start position:0% and the AMD OpenCL kernel currently required<00:19:02.410> additional<00:19:02.740> scratch<00:19:03.220> pad<00:19:03.460> space<00:19:03.730> in 00:19:04.080 --> 00:19:04.090 align:start position:0% required additional scratch pad space in 00:19:04.090 --> 00:19:07.260 align:start position:0% required additional scratch pad space in order<00:19:04.240> to<00:19:04.390> exchange<00:19:04.630> data<00:19:05.050> between<00:19:05.170> lanes<00:19:06.270> so 00:19:07.260 --> 00:19:07.270 align:start position:0% order to exchange data between lanes so 00:19:07.270 --> 00:19:09.930 align:start position:0% order to exchange data between lanes so in<00:19:07.390> order<00:19:07.540> to<00:19:07.780> effectively<00:19:08.280> execute<00:19:09.280> on<00:19:09.430> all 00:19:09.930 --> 00:19:09.940 align:start position:0% in order to effectively execute on all 00:19:09.940 --> 00:19:11.910 align:start position:0% in order to effectively execute on all existing<00:19:10.450> architectures<00:19:11.050> and<00:19:11.290> not<00:19:11.500> limit<00:19:11.800> our 00:19:11.910 --> 00:19:11.920 align:start position:0% existing architectures and not limit our 00:19:11.920 --> 00:19:14.250 align:start position:0% existing architectures and not limit our occupancy<00:19:12.310> the<00:19:13.120> case<00:19:13.330> portion<00:19:13.840> of<00:19:13.960> the<00:19:14.050> dag 00:19:14.250 --> 00:19:14.260 align:start position:0% occupancy the case portion of the dag 00:19:14.260 --> 00:19:17.580 align:start position:0% occupancy the case portion of the dag our<00:19:14.590> cash<00:19:15.550> bytes<00:19:15.850> up<00:19:16.060> there<00:19:16.270> is<00:19:16.450> set<00:19:17.050> to<00:19:17.170> 16 00:19:17.580 --> 00:19:17.590 align:start position:0% our cash bytes up there is set to 16 00:19:17.590 --> 00:19:18.710 align:start position:0% our cash bytes up there is set to 16 kilobytes 00:19:18.710 --> 00:19:18.720 align:start position:0% kilobytes 00:19:18.720 --> 00:19:22.260 align:start position:0% kilobytes now<00:19:19.720> the<00:19:20.260> compute<00:19:20.620> core<00:19:20.800> of<00:19:20.950> a<00:19:21.040> GPU<00:19:21.340> card<00:19:21.700> the 00:19:22.260 --> 00:19:22.270 align:start position:0% now the compute core of a GPU card the 00:19:22.270 --> 00:19:25.740 align:start position:0% now the compute core of a GPU card the ASIC<00:19:22.660> itself<00:19:23.080> is<00:19:23.880> really<00:19:24.880> just<00:19:25.270> a<00:19:25.480> large 00:19:25.740 --> 00:19:25.750 align:start position:0% ASIC itself is really just a large 00:19:25.750 --> 00:19:28.830 align:start position:0% ASIC itself is really just a large number<00:19:26.170> of<00:19:26.320> registers<00:19:26.920> that<00:19:27.840> feed 00:19:28.830 --> 00:19:28.840 align:start position:0% number of registers that feed 00:19:28.840 --> 00:19:32.060 align:start position:0% number of registers that feed high-throughput<00:19:29.730> programmable<00:19:30.730> math<00:19:31.150> units 00:19:32.060 --> 00:19:32.070 align:start position:0% high-throughput programmable math units 00:19:32.070 --> 00:19:35.700 align:start position:0% high-throughput programmable math units the<00:19:33.070> inner<00:19:33.250> loop<00:19:33.520> of<00:19:33.760> Eid<00:19:33.940> hash<00:19:34.120> just<00:19:35.050> uses<00:19:35.530> the 00:19:35.700 --> 00:19:35.710 align:start position:0% the inner loop of Eid hash just uses the 00:19:35.710 --> 00:19:38.880 align:start position:0% the inner loop of Eid hash just uses the dag<00:19:35.860> load<00:19:36.160> and<00:19:36.430> then<00:19:36.880> the<00:19:37.140> FNV<00:19:38.140> to<00:19:38.380> merge<00:19:38.620> the 00:19:38.880 --> 00:19:38.890 align:start position:0% dag load and then the FNV to merge the 00:19:38.890 --> 00:19:42.870 align:start position:0% dag load and then the FNV to merge the data<00:19:39.040> into<00:19:39.670> a<00:19:39.730> small<00:19:40.000> mixed<00:19:40.300> state<00:19:40.800> Prague<00:19:41.880> as 00:19:42.870 --> 00:19:42.880 align:start position:0% data into a small mixed state Prague as 00:19:42.880 --> 00:19:45.360 align:start position:0% data into a small mixed state Prague as a<00:19:43.390> sequence<00:19:43.960> of<00:19:44.110> random<00:19:44.290> math<00:19:44.680> instructions 00:19:45.360 --> 00:19:45.370 align:start position:0% a sequence of random math instructions 00:19:45.370 --> 00:19:47.700 align:start position:0% a sequence of random math instructions and<00:19:45.580> random<00:19:45.970> cache<00:19:46.210> reads<00:19:46.930> that<00:19:47.260> get<00:19:47.410> merged 00:19:47.700 --> 00:19:47.710 align:start position:0% and random cache reads that get merged 00:19:47.710 --> 00:19:50.490 align:start position:0% and random cache reads that get merged into<00:19:48.010> a<00:19:48.130> much<00:19:48.460> larger<00:19:49.090> mixed<00:19:49.360> state<00:19:49.540> why<00:19:50.440> is 00:19:50.490 --> 00:19:50.500 align:start position:0% into a much larger mixed state why is 00:19:50.500 --> 00:19:52.440 align:start position:0% into a much larger mixed state why is this<00:19:50.800> important<00:19:51.370> the<00:19:51.790> bigger<00:19:52.000> the<00:19:52.180> mixed 00:19:52.440 --> 00:19:52.450 align:start position:0% this important the bigger the mixed 00:19:52.450 --> 00:19:54.240 align:start position:0% this important the bigger the mixed state<00:19:52.630> and<00:19:52.840> the<00:19:53.020> more<00:19:53.200> random<00:19:53.410> the<00:19:53.860> bigger<00:19:54.040> the 00:19:54.240 --> 00:19:54.250 align:start position:0% state and the more random the bigger the 00:19:54.250 --> 00:19:56.310 align:start position:0% state and the more random the bigger the die<00:19:54.400> size<00:19:54.640> gets<00:19:55.030> on<00:19:55.450> fixed<00:19:55.930> functional 00:19:56.310 --> 00:19:56.320 align:start position:0% die size gets on fixed functional 00:19:56.320 --> 00:19:58.350 align:start position:0% die size gets on fixed functional hardware 00:19:58.350 --> 00:19:58.360 align:start position:0% hardware 00:19:58.360 --> 00:20:00.240 align:start position:0% hardware reeking<00:19:58.570> of<00:19:58.660> FNV<00:19:59.140> I<00:19:59.350> wanted<00:19:59.740> to<00:19:59.830> touch<00:19:59.950> on<00:20:00.130> it 00:20:00.240 --> 00:20:00.250 align:start position:0% reeking of FNV I wanted to touch on it 00:20:00.250 --> 00:20:02.280 align:start position:0% reeking of FNV I wanted to touch on it for<00:20:00.370> a<00:20:00.430> moment<00:20:00.760> because<00:20:01.390> I<00:20:01.540> did<00:20:01.750> promise<00:20:01.990> some 00:20:02.280 --> 00:20:02.290 align:start position:0% for a moment because I did promise some 00:20:02.290 --> 00:20:07.230 align:start position:0% for a moment because I did promise some folks<00:20:02.500> that<00:20:02.890> I<00:20:02.920> would<00:20:05.220> so<00:20:06.220> there's<00:20:06.610> this<00:20:06.880> rumor 00:20:07.230 --> 00:20:07.240 align:start position:0% folks that I would so there's this rumor 00:20:07.240 --> 00:20:09.690 align:start position:0% folks that I would so there's this rumor a<00:20:07.270> myth<00:20:07.540> going<00:20:07.570> around<00:20:08.110> about<00:20:08.470> FPGAs<00:20:09.430> with 00:20:09.690 --> 00:20:09.700 align:start position:0% a myth going around about FPGAs with 00:20:09.700 --> 00:20:12.270 align:start position:0% a myth going around about FPGAs with some<00:20:09.880> killer<00:20:10.150> performance<00:20:10.360> on<00:20:10.810> it<00:20:10.960> hash<00:20:11.280> even 00:20:12.270 --> 00:20:12.280 align:start position:0% some killer performance on it hash even 00:20:12.280 --> 00:20:14.930 align:start position:0% some killer performance on it hash even when<00:20:12.460> they're<00:20:12.640> just<00:20:12.820> bound<00:20:13.000> to<00:20:13.320> ddr4<00:20:14.320> memory 00:20:14.930 --> 00:20:14.940 align:start position:0% when they're just bound to ddr4 memory 00:20:14.940 --> 00:20:19.200 align:start position:0% when they're just bound to ddr4 memory F&V<00:20:15.940> fuller<00:20:16.600> Nova<00:20:17.020> vole<00:20:17.410> is<00:20:17.760> not<00:20:18.760> a<00:20:18.790> secure 00:20:19.200 --> 00:20:19.210 align:start position:0% F&V fuller Nova vole is not a secure 00:20:19.210 --> 00:20:22.200 align:start position:0% F&V fuller Nova vole is not a secure hash<00:20:19.390> function<00:20:20.430> there<00:20:21.430> is<00:20:21.460> a<00:20:21.610> possible<00:20:22.090> attack 00:20:22.200 --> 00:20:22.210 align:start position:0% hash function there is a possible attack 00:20:22.210 --> 00:20:24.830 align:start position:0% hash function there is a possible attack where<00:20:22.960> if<00:20:23.290> you<00:20:23.470> know<00:20:23.620> iteration<00:20:24.370> I<00:20:24.550> is 00:20:24.830 --> 00:20:24.840 align:start position:0% where if you know iteration I is 00:20:24.840 --> 00:20:29.220 align:start position:0% where if you know iteration I is accessing<00:20:25.840> dag<00:20:26.050> X<00:20:26.580> then<00:20:27.580> only<00:20:28.330> a<00:20:28.630> small<00:20:29.020> number 00:20:29.220 --> 00:20:29.230 align:start position:0% accessing dag X then only a small number 00:20:29.230 --> 00:20:31.620 align:start position:0% accessing dag X then only a small number of<00:20:29.380> next<00:20:29.830> possible<00:20:30.250> dag<00:20:30.460> Y<00:20:30.760> will<00:20:31.090> be<00:20:31.180> accessed 00:20:31.620 --> 00:20:31.630 align:start position:0% of next possible dag Y will be accessed 00:20:31.630 --> 00:20:35.280 align:start position:0% of next possible dag Y will be accessed and<00:20:32.010> technically<00:20:33.010> you<00:20:33.550> can<00:20:33.730> do<00:20:33.880> a<00:20:34.290> pre 00:20:35.280 --> 00:20:35.290 align:start position:0% and technically you can do a pre 00:20:35.290 --> 00:20:38.640 align:start position:0% and technically you can do a pre calculation<00:20:35.950> attack<00:20:36.450> where<00:20:37.450> you<00:20:37.720> pre<00:20:38.230> compute 00:20:38.640 --> 00:20:38.650 align:start position:0% calculation attack where you pre compute 00:20:38.650 --> 00:20:40.650 align:start position:0% calculation attack where you pre compute dag<00:20:38.860> X<00:20:39.100> with<00:20:39.370> enough<00:20:39.610> SRAM<00:20:40.120> or<00:20:40.300> be<00:20:40.420> ram 00:20:40.650 --> 00:20:40.660 align:start position:0% dag X with enough SRAM or be ram 00:20:40.660 --> 00:20:42.840 align:start position:0% dag X with enough SRAM or be ram followed<00:20:41.350> by<00:20:41.470> all<00:20:41.620> combinations<00:20:42.310> of<00:20:42.430> dag<00:20:42.640> Y 00:20:42.840 --> 00:20:42.850 align:start position:0% followed by all combinations of dag Y 00:20:42.850 --> 00:20:45.120 align:start position:0% followed by all combinations of dag Y and<00:20:43.150> you<00:20:43.690> completely<00:20:44.110> skip<00:20:44.470> off<00:20:44.650> the<00:20:44.830> memory 00:20:45.120 --> 00:20:45.130 align:start position:0% and you completely skip off the memory 00:20:45.130 --> 00:20:47.490 align:start position:0% and you completely skip off the memory accesses<00:20:45.580> what<00:20:46.330> that<00:20:46.510> means<00:20:46.780> is<00:20:47.110> that<00:20:47.140> you 00:20:47.490 --> 00:20:47.500 align:start position:0% accesses what that means is that you 00:20:47.500 --> 00:20:50.210 align:start position:0% accesses what that means is that you violate<00:20:47.950> the<00:20:48.100> memory<00:20:48.400> bandwidth<00:20:48.850> Evette<00:20:49.420> hash 00:20:50.210 --> 00:20:50.220 align:start position:0% violate the memory bandwidth Evette hash 00:20:50.220 --> 00:20:53.100 align:start position:0% violate the memory bandwidth Evette hash most<00:20:51.220> of<00:20:51.400> these<00:20:51.490> FPGA<00:20:52.150> rumors<00:20:52.660> seem<00:20:52.930> to<00:20:53.020> be 00:20:53.100 --> 00:20:53.110 align:start position:0% most of these FPGA rumors seem to be 00:20:53.110 --> 00:20:55.530 align:start position:0% most of these FPGA rumors seem to be require<00:20:53.410> 64<00:20:54.040> gigabyte<00:20:54.670> which<00:20:54.700> suggests 00:20:55.530 --> 00:20:55.540 align:start position:0% require 64 gigabyte which suggests 00:20:55.540 --> 00:20:58.350 align:start position:0% require 64 gigabyte which suggests pre-computation<00:20:56.550> combining<00:20:57.550> DAGs<00:20:57.790> doesn't 00:20:58.350 --> 00:20:58.360 align:start position:0% pre-computation combining DAGs doesn't 00:20:58.360 --> 00:21:01.260 align:start position:0% pre-computation combining DAGs doesn't actually<00:20:58.510> work<00:20:59.040> but<00:21:00.040> in<00:21:00.430> case<00:21:00.760> there<00:21:00.970> is<00:21:01.000> an 00:21:01.260 --> 00:21:01.270 align:start position:0% actually work but in case there is an 00:21:01.270 --> 00:21:03.750 align:start position:0% actually work but in case there is an F&V<00:21:01.540> attack<00:21:02.050> well<00:21:02.770> there's<00:21:03.280> kind<00:21:03.460> of<00:21:03.580> good 00:21:03.750 --> 00:21:03.760 align:start position:0% F&V attack well there's kind of good 00:21:03.760 --> 00:21:04.290 align:start position:0% F&V attack well there's kind of good news 00:21:04.290 --> 00:21:04.300 align:start position:0% news 00:21:04.300 --> 00:21:07.410 align:start position:0% news see<00:21:04.720> prog<00:21:05.500> pal<00:21:05.740> doesn't<00:21:06.610> have<00:21:06.910> this<00:21:07.150> problem 00:21:07.410 --> 00:21:07.420 align:start position:0% see prog pal doesn't have this problem 00:21:07.420 --> 00:21:12.540 align:start position:0% see prog pal doesn't have this problem and<00:21:07.990> hash<00:21:08.830> is<00:21:09.510> dag<00:21:10.510> to<00:21:10.990> F<00:21:11.140> and<00:21:11.350> V<00:21:11.380> to<00:21:11.680> dag<00:21:11.890> but 00:21:12.540 --> 00:21:12.550 align:start position:0% and hash is dag to F and V to dag but 00:21:12.550 --> 00:21:15.840 align:start position:0% and hash is dag to F and V to dag but prog<00:21:13.090> pal<00:21:13.500> its<00:21:14.500> dag<00:21:14.830> with<00:21:15.250> a<00:21:15.280> bunch<00:21:15.550> of<00:21:15.640> random 00:21:15.840 --> 00:21:15.850 align:start position:0% prog pal its dag with a bunch of random 00:21:15.850 --> 00:21:17.910 align:start position:0% prog pal its dag with a bunch of random math<00:21:16.150> which<00:21:16.540> with<00:21:16.960> dag<00:21:17.230> with<00:21:17.530> a<00:21:17.560> bunch<00:21:17.800> of 00:21:17.910 --> 00:21:17.920 align:start position:0% math which with dag with a bunch of 00:21:17.920 --> 00:21:20.190 align:start position:0% math which with dag with a bunch of random<00:21:18.100> masks<00:21:18.400> repeat<00:21:19.090> repeat<00:21:19.300> repeat<00:21:19.510> until 00:21:20.190 --> 00:21:20.200 align:start position:0% random masks repeat repeat repeat until 00:21:20.200 --> 00:21:24.780 align:start position:0% random masks repeat repeat repeat until dag<00:21:21.750> now<00:21:22.750> multiple<00:21:23.740> people<00:21:24.040> have<00:21:24.130> claimed<00:21:24.490> to 00:21:24.780 --> 00:21:24.790 align:start position:0% dag now multiple people have claimed to 00:21:24.790 --> 00:21:26.910 align:start position:0% dag now multiple people have claimed to attack<00:21:25.240> this<00:21:25.510> weakness<00:21:25.780> and<00:21:26.140> F<00:21:26.230> and<00:21:26.410> V<00:21:26.470> to<00:21:26.740> date 00:21:26.910 --> 00:21:26.920 align:start position:0% attack this weakness and F and V to date 00:21:26.920 --> 00:21:32.670 align:start position:0% attack this weakness and F and V to date and<00:21:27.190> who<00:21:28.000> knows<00:21:28.180> maybe<00:21:28.480> I<00:21:28.960> might<00:21:29.260> have<00:21:29.410> to<00:21:31.680> so 00:21:32.670 --> 00:21:32.680 align:start position:0% and who knows maybe I might have to so 00:21:32.680 --> 00:21:34.140 align:start position:0% and who knows maybe I might have to so let's<00:21:33.100> move<00:21:33.280> on<00:21:33.460> to<00:21:33.490> the<00:21:33.670> random<00:21:33.970> math 00:21:34.140 --> 00:21:34.150 align:start position:0% let's move on to the random math 00:21:34.150 --> 00:21:36.560 align:start position:0% let's move on to the random math function<00:21:34.600> and<00:21:34.720> why<00:21:34.810> that<00:21:34.960> matters<00:21:35.140> so<00:21:35.350> much 00:21:36.560 --> 00:21:36.570 align:start position:0% function and why that matters so much 00:21:36.570 --> 00:21:39.480 align:start position:0% function and why that matters so much randomness<00:21:37.570> and<00:21:37.840> fixed<00:21:38.410> function<00:21:38.740> Hardware 00:21:39.480 --> 00:21:39.490 align:start position:0% randomness and fixed function Hardware 00:21:39.490 --> 00:21:43.830 align:start position:0% randomness and fixed function Hardware do<00:21:40.090> not<00:21:40.390> mix<00:21:41.550> random<00:21:42.550> math<00:21:42.820> helps<00:21:43.540> to<00:21:43.660> match 00:21:43.830 --> 00:21:43.840 align:start position:0% do not mix random math helps to match 00:21:43.840 --> 00:21:46.590 align:start position:0% do not mix random math helps to match the<00:21:44.290> GPUs<00:21:44.740> program<00:21:45.220> ability<00:21:45.730> and<00:21:45.910> integer 00:21:46.590 --> 00:21:46.600 align:start position:0% the GPUs program ability and integer 00:21:46.600 --> 00:21:49.860 align:start position:0% the GPUs program ability and integer math<00:21:46.780> that<00:21:47.650> they<00:21:47.890> normally<00:21:48.040> excel<00:21:48.640> at<00:21:48.870> fixed 00:21:49.860 --> 00:21:49.870 align:start position:0% math that they normally excel at fixed 00:21:49.870 --> 00:21:51.960 align:start position:0% math that they normally excel at fixed function<00:21:50.200> Hardware<00:21:50.800> like<00:21:51.160> I<00:21:51.190> said<00:21:51.490> doesn't 00:21:51.960 --> 00:21:51.970 align:start position:0% function Hardware like I said doesn't 00:21:51.970 --> 00:21:54.420 align:start position:0% function Hardware like I said doesn't like<00:21:52.060> randomness<00:21:52.770> FPGAs<00:21:53.770> don't<00:21:54.040> really<00:21:54.220> like 00:21:54.420 --> 00:21:54.430 align:start position:0% like randomness FPGAs don't really like 00:21:54.430 --> 00:21:55.230 align:start position:0% like randomness FPGAs don't really like it<00:21:54.580> much<00:21:54.730> either 00:21:55.230 --> 00:21:55.240 align:start position:0% it much either 00:21:55.240 --> 00:21:58.080 align:start position:0% it much either they<00:21:55.840> require<00:21:56.200> reprogram<00:21:56.920> ability<00:21:57.370> via<00:21:57.550> JTAG 00:21:58.080 --> 00:21:58.090 align:start position:0% they require reprogram ability via JTAG 00:21:58.090 --> 00:22:00.060 align:start position:0% they require reprogram ability via JTAG USB<00:21:58.630> or<00:21:58.810> some<00:21:59.020> other<00:21:59.170> metric<00:21:59.650> to<00:21:59.770> employ 00:22:00.060 --> 00:22:00.070 align:start position:0% USB or some other metric to employ 00:22:00.070 --> 00:22:02.820 align:start position:0% USB or some other metric to employ randomness<00:22:00.750> some<00:22:01.750> people<00:22:02.110> think<00:22:02.440> that<00:22:02.500> the 00:22:02.820 --> 00:22:02.830 align:start position:0% randomness some people think that the 00:22:02.830 --> 00:22:05.070 align:start position:0% randomness some people think that the random<00:22:03.550> math<00:22:03.730> and<00:22:03.970> probe<00:22:04.150> pal<00:22:04.390> is<00:22:04.630> a<00:22:04.660> bit<00:22:04.870> naive 00:22:05.070 --> 00:22:05.080 align:start position:0% random math and probe pal is a bit naive 00:22:05.080 --> 00:22:07.830 align:start position:0% random math and probe pal is a bit naive they<00:22:05.680> assume<00:22:06.490> it's<00:22:06.730> cheap<00:22:07.000> to<00:22:07.180> implement<00:22:07.600> on<00:22:07.690> a 00:22:07.830 --> 00:22:07.840 align:start position:0% they assume it's cheap to implement on a 00:22:07.840 --> 00:22:10.700 align:start position:0% they assume it's cheap to implement on a 6<00:22:08.140> because<00:22:08.800> we<00:22:08.830> use<00:22:09.100> bit<00:22:09.370> operations 00:22:10.700 --> 00:22:10.710 align:start position:0% 6 because we use bit operations 00:22:10.710 --> 00:22:13.070 align:start position:0% 6 because we use bit operations our<00:22:11.160> implementation<00:22:11.970> uses<00:22:12.270> rotates<00:22:12.660> which 00:22:13.070 --> 00:22:13.080 align:start position:0% our implementation uses rotates which 00:22:13.080 --> 00:22:15.080 align:start position:0% our implementation uses rotates which require<00:22:13.440> a<00:22:13.470> full<00:22:13.800> barrel<00:22:14.040> shifter<00:22:14.610> that's 00:22:15.080 --> 00:22:15.090 align:start position:0% require a full barrel shifter that's 00:22:15.090 --> 00:22:17.030 align:start position:0% require a full barrel shifter that's pretty<00:22:15.570> expensive<00:22:15.870> on<00:22:16.380> fixed<00:22:16.890> function 00:22:17.030 --> 00:22:17.040 align:start position:0% pretty expensive on fixed function 00:22:17.040 --> 00:22:21.170 align:start position:0% pretty expensive on fixed function hardware<00:22:17.610> and<00:22:18.980> sure<00:22:19.980> if<00:22:20.520> you're<00:22:20.820> the 00:22:21.170 --> 00:22:21.180 align:start position:0% hardware and sure if you're the 00:22:21.180 --> 00:22:23.060 align:start position:0% hardware and sure if you're the bitstream<00:22:21.480> on<00:22:21.780> an<00:22:21.870> FPGA<00:22:22.380> could<00:22:22.740> be<00:22:22.890> updated 00:22:23.060 --> 00:22:23.070 align:start position:0% bitstream on an FPGA could be updated 00:22:23.070 --> 00:22:25.820 align:start position:0% bitstream on an FPGA could be updated every<00:22:23.520> 12<00:22:23.940> minutes<00:22:24.230> it<00:22:25.230> could<00:22:25.530> definitely 00:22:25.820 --> 00:22:25.830 align:start position:0% every 12 minutes it could definitely 00:22:25.830 --> 00:22:28.070 align:start position:0% every 12 minutes it could definitely implement<00:22:26.580> the<00:22:26.640> random<00:22:27.270> math<00:22:27.510> portion<00:22:27.810> of 00:22:28.070 --> 00:22:28.080 align:start position:0% implement the random math portion of 00:22:28.080 --> 00:22:30.830 align:start position:0% implement the random math portion of prog<00:22:28.290> pal<00:22:28.530> at<00:22:28.800> a<00:22:28.830> lower<00:22:29.070> power<00:22:29.340> than<00:22:30.060> GPU<00:22:30.630> by 00:22:30.830 --> 00:22:30.840 align:start position:0% prog pal at a lower power than GPU by 00:22:30.840 --> 00:22:32.540 align:start position:0% prog pal at a lower power than GPU by creating<00:22:31.290> a<00:22:31.350> fixed<00:22:31.650> pipeline<00:22:32.100> but<00:22:32.310> there's<00:22:32.490> a 00:22:32.540 --> 00:22:32.550 align:start position:0% creating a fixed pipeline but there's a 00:22:32.550 --> 00:22:34.580 align:start position:0% creating a fixed pipeline but there's a problem<00:22:32.820> with<00:22:33.090> that<00:22:33.270> the<00:22:33.750> cache<00:22:33.990> accesses 00:22:34.580 --> 00:22:34.590 align:start position:0% problem with that the cache accesses 00:22:34.590 --> 00:22:37.760 align:start position:0% problem with that the cache accesses however<00:22:35.490> all<00:22:35.670> of<00:22:35.820> that<00:22:35.970> all<00:22:36.140> so<00:22:37.140> since 00:22:37.760 --> 00:22:37.770 align:start position:0% however all of that all so since 00:22:37.770 --> 00:22:38.780 align:start position:0% however all of that all so since you're<00:22:37.920> gonna<00:22:38.010> need<00:22:38.220> a<00:22:38.280> high<00:22:38.490> throughput 00:22:38.780 --> 00:22:38.790 align:start position:0% you're gonna need a high throughput 00:22:38.790 --> 00:22:42.290 align:start position:0% you're gonna need a high throughput throughput<00:22:39.750> crossbar<00:22:40.790> you<00:22:41.790> know<00:22:41.910> the<00:22:42.090> things 00:22:42.290 --> 00:22:42.300 align:start position:0% throughput crossbar you know the things 00:22:42.300 --> 00:22:44.780 align:start position:0% throughput crossbar you know the things GPUs<00:22:42.810> naturally<00:22:43.290> have<00:22:43.500> it's<00:22:44.370> gonna<00:22:44.490> cause<00:22:44.760> a 00:22:44.780 --> 00:22:44.790 align:start position:0% GPUs naturally have it's gonna cause a 00:22:44.790 --> 00:22:48.320 align:start position:0% GPUs naturally have it's gonna cause a problem<00:22:45.060> and<00:22:45.680> we<00:22:46.680> use<00:22:46.920> KISS<00:22:47.190> 99<00:22:47.730> for<00:22:48.090> our 00:22:48.320 --> 00:22:48.330 align:start position:0% problem and we use KISS 99 for our 00:22:48.330 --> 00:22:50.570 align:start position:0% problem and we use KISS 99 for our random<00:22:48.810> random<00:22:49.530> math<00:22:49.740> and<00:22:49.980> that's<00:22:50.130> that's 00:22:50.570 --> 00:22:50.580 align:start position:0% random random math and that's that's 00:22:50.580 --> 00:22:52.700 align:start position:0% random random math and that's that's kind<00:22:50.790> of<00:22:51.030> important<00:22:51.540> for<00:22:51.750> you<00:22:51.840> number<00:22:52.440> geeks 00:22:52.700 --> 00:22:52.710 align:start position:0% kind of important for you number geeks 00:22:52.710 --> 00:22:54.920 align:start position:0% kind of important for you number geeks out<00:22:52.920> there<00:22:53.130> like<00:22:53.340> me<00:22:53.550> we<00:22:54.030> don't<00:22:54.240> use<00:22:54.420> Mersenne 00:22:54.920 --> 00:22:54.930 align:start position:0% out there like me we don't use Mersenne 00:22:54.930 --> 00:22:57.200 align:start position:0% out there like me we don't use Mersenne twister<00:22:55.080> because<00:22:56.070> it's<00:22:56.280> efficiently 00:22:57.200 --> 00:22:57.210 align:start position:0% twister because it's efficiently 00:22:57.210 --> 00:22:58.910 align:start position:0% twister because it's efficiently implemented<00:22:57.570> on<00:22:57.840> the<00:22:57.960> specialized<00:22:58.440> ASIC 00:22:58.910 --> 00:22:58.920 align:start position:0% implemented on the specialized ASIC 00:22:58.920 --> 00:23:00.710 align:start position:0% implemented on the specialized ASIC actually<00:22:59.730> we<00:22:59.880> don't<00:23:00.060> use<00:23:00.240> a<00:23:00.270> lot<00:23:00.390> of<00:23:00.570> the 00:23:00.710 --> 00:23:00.720 align:start position:0% actually we don't use a lot of the 00:23:00.720 --> 00:23:02.240 align:start position:0% actually we don't use a lot of the random<00:23:01.050> number<00:23:01.290> generators<00:23:01.800> out<00:23:02.040> there 00:23:02.240 --> 00:23:02.250 align:start position:0% random number generators out there 00:23:02.250 --> 00:23:05.000 align:start position:0% random number generators out there because<00:23:02.880> we<00:23:03.600> need<00:23:03.780> something<00:23:04.200> that<00:23:04.230> posts<00:23:04.830> the 00:23:05.000 --> 00:23:05.010 align:start position:0% because we need something that posts the 00:23:05.010 --> 00:23:07.760 align:start position:0% because we need something that posts the test<00:23:05.250> you<00:23:05.640> owe<00:23:05.670> one<00:23:06.330> statistical<00:23:07.200> test<00:23:07.530> suite 00:23:07.760 --> 00:23:07.770 align:start position:0% test you owe one statistical test suite 00:23:07.770 --> 00:23:12.470 align:start position:0% test you owe one statistical test suite why<00:23:08.210> well<00:23:10.070> test<00:23:11.070> you<00:23:11.280> oh<00:23:11.310> one<00:23:11.640> actually 00:23:12.470 --> 00:23:12.480 align:start position:0% why well test you oh one actually 00:23:12.480 --> 00:23:14.540 align:start position:0% why well test you oh one actually performs<00:23:13.110> checks<00:23:13.500> to<00:23:13.710> determine<00:23:13.950> whether<00:23:14.430> a 00:23:14.540 --> 00:23:14.550 align:start position:0% performs checks to determine whether a 00:23:14.550 --> 00:23:16.400 align:start position:0% performs checks to determine whether a random<00:23:15.000> number<00:23:15.120> generator<00:23:15.330> produces<00:23:15.930> well 00:23:16.400 --> 00:23:16.410 align:start position:0% random number generator produces well 00:23:16.410 --> 00:23:19.960 align:start position:0% random number generator produces well truly<00:23:17.250> random<00:23:17.550> numbers<00:23:17.910> for<00:23:18.660> some<00:23:18.870> reason 00:23:19.960 --> 00:23:19.970 align:start position:0% truly random numbers for some reason 00:23:19.970 --> 00:23:24.860 align:start position:0% truly random numbers for some reason most<00:23:20.970> of<00:23:21.120> them<00:23:21.270> fail<00:23:21.570> horribly 00:23:24.860 --> 00:23:24.870 align:start position:0% 00:23:24.870 --> 00:23:28.520 align:start position:0% so<00:23:25.380> yeah<00:23:26.130> Zor<00:23:26.670> shifts<00:23:27.180> they're<00:23:28.050> not<00:23:28.380> your 00:23:28.520 --> 00:23:28.530 align:start position:0% so yeah Zor shifts they're not your 00:23:28.530 --> 00:23:31.550 align:start position:0% so yeah Zor shifts they're not your friend<00:23:28.800> folks<00:23:29.130> they're<00:23:29.580> not<00:23:29.760> your<00:23:29.880> friend<00:23:30.560> so 00:23:31.550 --> 00:23:31.560 align:start position:0% friend folks they're not your friend so 00:23:31.560 --> 00:23:33.470 align:start position:0% friend folks they're not your friend so by<00:23:32.160> now<00:23:32.310> you're<00:23:32.430> getting<00:23:32.700> kind<00:23:32.880> of<00:23:33.030> bored<00:23:33.300> and 00:23:33.470 --> 00:23:33.480 align:start position:0% by now you're getting kind of bored and 00:23:33.480 --> 00:23:35.480 align:start position:0% by now you're getting kind of bored and you're<00:23:33.600> thinking<00:23:33.810> okay<00:23:34.290> okay<00:23:34.620> great<00:23:34.920> you 00:23:35.480 --> 00:23:35.490 align:start position:0% you're thinking okay okay great you 00:23:35.490 --> 00:23:37.040 align:start position:0% you're thinking okay okay great you talked<00:23:35.730> a<00:23:35.820> little<00:23:35.910> bit<00:23:36.090> about<00:23:36.240> Asics<00:23:36.750> what<00:23:36.960> are 00:23:37.040 --> 00:23:37.050 align:start position:0% talked a little bit about Asics what are 00:23:37.050 --> 00:23:41.060 align:start position:0% talked a little bit about Asics what are the<00:23:37.140> actual<00:23:37.500> results<00:23:38.040> well<00:23:39.290> boom<00:23:40.290> baby 00:23:41.060 --> 00:23:41.070 align:start position:0% the actual results well boom baby 00:23:41.070 --> 00:23:44.540 align:start position:0% the actual results well boom baby look<00:23:41.490> at<00:23:41.580> that<00:23:42.020> look<00:23:43.020> at<00:23:43.140> that<00:23:43.320> almost<00:23:43.740> 90% 00:23:44.540 --> 00:23:44.550 align:start position:0% look at that look at that almost 90% 00:23:44.550 --> 00:23:46.520 align:start position:0% look at that look at that almost 90% utilization<00:23:45.030> in<00:23:45.810> both<00:23:45.960> the<00:23:46.110> core<00:23:46.320> and<00:23:46.440> the 00:23:46.520 --> 00:23:46.530 align:start position:0% utilization in both the core and the 00:23:46.530 --> 00:23:49.850 align:start position:0% utilization in both the core and the memory<00:23:46.860> and<00:23:47.600> on<00:23:48.600> top<00:23:48.870> of<00:23:48.990> that<00:23:49.110> we<00:23:49.560> can<00:23:49.590> also 00:23:49.850 --> 00:23:49.860 align:start position:0% memory and on top of that we can also 00:23:49.860 --> 00:23:52.190 align:start position:0% memory and on top of that we can also see<00:23:50.130> our<00:23:50.160> scratchpad<00:23:50.580> memory<00:23:50.820> up<00:23:51.240> there<00:23:51.480> the 00:23:52.190 --> 00:23:52.200 align:start position:0% see our scratchpad memory up there the 00:23:52.200 --> 00:23:55.340 align:start position:0% see our scratchpad memory up there the second<00:23:52.890> screen<00:23:53.100> is<00:23:53.870> also<00:23:54.870> completely 00:23:55.340 --> 00:23:55.350 align:start position:0% second screen is also completely 00:23:55.350 --> 00:23:59.480 align:start position:0% second screen is also completely saturated<00:23:55.980> and<00:23:56.250> then<00:23:56.700> look<00:23:56.850> at<00:23:56.940> the<00:23:57.030> 5<00:23:58.220> 85<00:23:59.220> 80 00:23:59.480 --> 00:23:59.490 align:start position:0% saturated and then look at the 5 85 80 00:23:59.490 --> 00:24:02.830 align:start position:0% saturated and then look at the 5 85 80 has<00:24:00.650> 88.8% 00:24:02.830 --> 00:24:02.840 align:start position:0% has 88.8% 00:24:02.840 --> 00:24:06.410 align:start position:0% has 88.8% occupancy<00:24:03.840> and<00:24:04.170> saturation<00:24:05.040> up<00:24:05.610> there<00:24:06.150> on<00:24:06.360> a 00:24:06.410 --> 00:24:06.420 align:start position:0% occupancy and saturation up there on a 00:24:06.420 --> 00:24:09.670 align:start position:0% occupancy and saturation up there on a video<00:24:06.780> eighty<00:24:07.380> eight<00:24:07.680> point<00:24:07.920> four<00:24:08.280> percent 00:24:09.670 --> 00:24:09.680 align:start position:0% video eighty eight point four percent 00:24:09.680 --> 00:24:12.710 align:start position:0% video eighty eight point four percent now<00:24:10.680> in<00:24:10.830> at<00:24:11.010> hash<00:24:11.280> we<00:24:11.790> all<00:24:11.910> know<00:24:12.120> that<00:24:12.300> NVIDIA 00:24:12.710 --> 00:24:12.720 align:start position:0% now in at hash we all know that NVIDIA 00:24:12.720 --> 00:24:15.440 align:start position:0% now in at hash we all know that NVIDIA GPUs<00:24:13.170> and<00:24:13.380> AMD<00:24:13.710> GPUs<00:24:14.220> are<00:24:14.400> not<00:24:14.670> equal<00:24:15.090> they're 00:24:15.440 --> 00:24:15.450 align:start position:0% GPUs and AMD GPUs are not equal they're 00:24:15.450 --> 00:24:19.250 align:start position:0% GPUs and AMD GPUs are not equal they're not<00:24:15.630> matched<00:24:16.520> they're<00:24:17.520> not<00:24:17.670> balanced<00:24:18.260> prog 00:24:19.250 --> 00:24:19.260 align:start position:0% not matched they're not balanced prog 00:24:19.260 --> 00:24:21.110 align:start position:0% not matched they're not balanced prog power<00:24:19.470> fixes<00:24:19.890> that 00:24:21.110 --> 00:24:21.120 align:start position:0% power fixes that 00:24:21.120 --> 00:24:23.570 align:start position:0% power fixes that a<00:24:21.180> proper<00:24:21.690> proof-of-work<00:24:22.290> algorithm<00:24:22.920> is<00:24:23.100> all 00:24:23.570 --> 00:24:23.580 align:start position:0% a proper proof-of-work algorithm is all 00:24:23.580 --> 00:24:29.350 align:start position:0% a proper proof-of-work algorithm is all about<00:24:24.120> balance<00:24:24.750> and<00:24:28.070> yeah 00:24:29.350 --> 00:24:29.360 align:start position:0% about balance and yeah 00:24:29.360 --> 00:24:32.540 align:start position:0% about balance and yeah while<00:24:30.360> our<00:24:30.570> cat<00:24:31.200> check<00:24:31.460> calculation<00:24:32.460> is 00:24:32.540 --> 00:24:32.550 align:start position:0% while our cat check calculation is 00:24:32.550 --> 00:24:35.120 align:start position:0% while our cat check calculation is halved<00:24:32.790> frog<00:24:33.750> power<00:24:33.990> does<00:24:34.230> add<00:24:34.470> that<00:24:34.500> series 00:24:35.120 --> 00:24:35.130 align:start position:0% halved frog power does add that series 00:24:35.130 --> 00:24:37.640 align:start position:0% halved frog power does add that series of<00:24:35.280> KISS<00:24:35.460> 99<00:24:35.970> calculations<00:24:36.750> as<00:24:37.170> part<00:24:37.470> of<00:24:37.559> our 00:24:37.640 --> 00:24:37.650 align:start position:0% of KISS 99 calculations as part of our 00:24:37.650 --> 00:24:39.890 align:start position:0% of KISS 99 calculations as part of our mixed<00:24:37.980> stage<00:24:38.160> remember<00:24:39.150> we<00:24:39.270> talked<00:24:39.510> about<00:24:39.750> why 00:24:39.890 --> 00:24:39.900 align:start position:0% mixed stage remember we talked about why 00:24:39.900 --> 00:24:41.000 align:start position:0% mixed stage remember we talked about why mixing<00:24:40.320> is<00:24:40.410> important 00:24:41.000 --> 00:24:41.010 align:start position:0% mixing is important 00:24:41.010 --> 00:24:43.220 align:start position:0% mixing is important well<00:24:41.520> the<00:24:41.790> filmic<00:24:42.150> stage<00:24:42.420> has<00:24:42.630> too<00:24:42.809> much<00:24:42.960> data 00:24:43.220 --> 00:24:43.230 align:start position:0% well the filmic stage has too much data 00:24:43.230 --> 00:24:45.500 align:start position:0% well the filmic stage has too much data to<00:24:43.410> offload<00:24:43.740> to<00:24:43.950> an<00:24:44.010> external<00:24:44.340> FPGA<00:24:44.970> now<00:24:45.240> it 00:24:45.500 --> 00:24:45.510 align:start position:0% to offload to an external FPGA now it 00:24:45.510 --> 00:24:47.810 align:start position:0% to offload to an external FPGA now it could<00:24:46.170> be<00:24:46.200> implemented<00:24:46.920> on<00:24:47.010> a<00:24:47.070> really<00:24:47.760> small 00:24:47.810 --> 00:24:47.820 align:start position:0% could be implemented on a really small 00:24:47.820 --> 00:24:49.760 align:start position:0% could be implemented on a really small chip<00:24:48.240> within<00:24:48.540> an<00:24:48.660> ASIC<00:24:49.020> using<00:24:49.380> a<00:24:49.440> small 00:24:49.760 --> 00:24:49.770 align:start position:0% chip within an ASIC using a small 00:24:49.770 --> 00:24:53.330 align:start position:0% chip within an ASIC using a small accelerator<00:24:50.340> but<00:24:51.059> that's<00:24:51.210> only<00:24:51.420> 7%<00:24:52.170> of<00:24:52.860> our 00:24:53.330 --> 00:24:53.340 align:start position:0% accelerator but that's only 7% of our 00:24:53.340 --> 00:24:56.630 align:start position:0% accelerator but that's only 7% of our compute<00:24:53.730> utilization<00:24:54.800> so<00:24:55.800> we<00:24:56.130> see<00:24:56.309> up<00:24:56.460> there 00:24:56.630 --> 00:24:56.640 align:start position:0% compute utilization so we see up there 00:24:56.640 --> 00:24:59.330 align:start position:0% compute utilization so we see up there ninety<00:24:57.450> one<00:24:57.690> point<00:24:58.020> nine<00:24:58.410> percent<00:24:58.440> of<00:24:59.130> the 00:24:59.330 --> 00:24:59.340 align:start position:0% ninety one point nine percent of the 00:24:59.340 --> 00:25:01.790 align:start position:0% ninety one point nine percent of the instructions<00:24:59.940> are<00:25:00.179> executed<00:25:00.500> within<00:25:01.500> the<00:25:01.620> dag 00:25:01.790 --> 00:25:01.800 align:start position:0% instructions are executed within the dag 00:25:01.800 --> 00:25:03.740 align:start position:0% instructions are executed within the dag access<00:25:02.429> random<00:25:02.880> math<00:25:03.059> and<00:25:03.270> random<00:25:03.510> cache 00:25:03.740 --> 00:25:03.750 align:start position:0% access random math and random cache 00:25:03.750 --> 00:25:06.140 align:start position:0% access random math and random cache accesses<00:25:04.710> things<00:25:05.280> we<00:25:05.429> can't<00:25:05.670> move<00:25:05.820> off<00:25:06.030> the 00:25:06.140 --> 00:25:06.150 align:start position:0% accesses things we can't move off the 00:25:06.150 --> 00:25:08.510 align:start position:0% accesses things we can't move off the GPU<00:25:06.540> can't<00:25:07.350> check<00:25:07.500> and<00:25:07.679> fill<00:25:07.830> mix<00:25:08.070> only 00:25:08.510 --> 00:25:08.520 align:start position:0% GPU can't check and fill mix only 00:25:08.520 --> 00:25:10.700 align:start position:0% GPU can't check and fill mix only account<00:25:09.000> for<00:25:09.059> about<00:25:09.179> seven<00:25:09.600> percent<00:25:09.710> that's 00:25:10.700 --> 00:25:10.710 align:start position:0% account for about seven percent that's 00:25:10.710 --> 00:25:14.210 align:start position:0% account for about seven percent that's only<00:25:10.950> a<00:25:11.010> seven<00:25:11.340> percent<00:25:11.580> speed<00:25:11.850> speed<00:25:12.570> gain<00:25:13.220> so 00:25:14.210 --> 00:25:14.220 align:start position:0% only a seven percent speed speed gain so 00:25:14.220 --> 00:25:16.040 align:start position:0% only a seven percent speed speed gain so what<00:25:14.850> does<00:25:15.090> that<00:25:15.270> actually<00:25:15.540> mean 00:25:16.040 --> 00:25:16.050 align:start position:0% what does that actually mean 00:25:16.050 --> 00:25:20.930 align:start position:0% what does that actually mean well<00:25:16.470> in<00:25:16.740> layman's<00:25:17.190> terms<00:25:17.630> it<00:25:18.630> means<00:25:19.700> that<00:25:20.700> an 00:25:20.930 --> 00:25:20.940 align:start position:0% well in layman's terms it means that an 00:25:20.940 --> 00:25:25.510 align:start position:0% well in layman's terms it means that an ASIC<00:25:21.360> for<00:25:21.540> at<00:25:21.660> hash<00:25:23.150> looks<00:25:24.150> just<00:25:24.900> like<00:25:25.080> this 00:25:25.510 --> 00:25:25.520 align:start position:0% ASIC for at hash looks just like this 00:25:25.520 --> 00:25:29.060 align:start position:0% ASIC for at hash looks just like this high<00:25:26.520> bandwidth<00:25:26.910> memory<00:25:27.150> interface<00:25:28.070> kept 00:25:29.060 --> 00:25:29.070 align:start position:0% high bandwidth memory interface kept 00:25:29.070 --> 00:25:31.910 align:start position:0% high bandwidth memory interface kept track<00:25:29.250> engine<00:25:29.670> usually<00:25:30.210> on<00:25:30.330> an<00:25:30.450> FPGA<00:25:30.960> and<00:25:31.320> a 00:25:31.910 --> 00:25:31.920 align:start position:0% track engine usually on an FPGA and a 00:25:31.920 --> 00:25:34.100 align:start position:0% track engine usually on an FPGA and a small<00:25:32.160> compute<00:25:32.580> core<00:25:32.820> to<00:25:32.970> do<00:25:33.090> FNV<00:25:33.570> loop<00:25:33.840> and 00:25:34.100 --> 00:25:34.110 align:start position:0% small compute core to do FNV loop and 00:25:34.110 --> 00:25:38.090 align:start position:0% small compute core to do FNV loop and modulo<00:25:34.530> operations<00:25:35.600> but<00:25:36.600> a<00:25:36.630> pro<00:25:36.900> palcic<00:25:37.650> well 00:25:38.090 --> 00:25:38.100 align:start position:0% modulo operations but a pro palcic well 00:25:38.100 --> 00:25:39.799 align:start position:0% modulo operations but a pro palcic well it's<00:25:38.790> gonna<00:25:38.940> have<00:25:39.030> the<00:25:39.179> high<00:25:39.390> bandwidth 00:25:39.799 --> 00:25:39.809 align:start position:0% it's gonna have the high bandwidth 00:25:39.809 --> 00:25:41.510 align:start position:0% it's gonna have the high bandwidth memory<00:25:39.929> interface<00:25:40.559> it's<00:25:41.160> going<00:25:41.309> to<00:25:41.370> have<00:25:41.490> a 00:25:41.510 --> 00:25:41.520 align:start position:0% memory interface it's going to have a 00:25:41.520 --> 00:25:43.960 align:start position:0% memory interface it's going to have a compute<00:25:41.910> core<00:25:42.150> with<00:25:42.450> a<00:25:42.480> large<00:25:42.840> register<00:25:43.530> file 00:25:43.960 --> 00:25:43.970 align:start position:0% compute core with a large register file 00:25:43.970 --> 00:25:45.760 align:start position:0% compute core with a large register file compute<00:25:44.970> core<00:25:45.240> with<00:25:45.450> high<00:25:45.720> throughput 00:25:45.760 --> 00:25:45.770 align:start position:0% compute core with high throughput 00:25:45.770 --> 00:25:48.919 align:start position:0% compute core with high throughput integer<00:25:46.770> math<00:25:47.070> high<00:25:47.970> throughput<00:25:48.330> highly 00:25:48.919 --> 00:25:48.929 align:start position:0% integer math high throughput highly 00:25:48.929 --> 00:25:51.980 align:start position:0% integer math high throughput highly banked<00:25:49.230> cache<00:25:49.640> small<00:25:50.640> cap<00:25:50.880> check<00:25:51.120> and<00:25:51.300> kiss<00:25:51.480> 99 00:25:51.980 --> 00:25:51.990 align:start position:0% banked cache small cap check and kiss 99 00:25:51.990 --> 00:25:55.220 align:start position:0% banked cache small cap check and kiss 99 engines<00:25:52.530> wait<00:25:53.190> I<00:25:53.400> think<00:25:54.270> that's<00:25:54.660> also<00:25:55.020> known 00:25:55.220 --> 00:25:55.230 align:start position:0% engines wait I think that's also known 00:25:55.230 --> 00:26:01.130 align:start position:0% engines wait I think that's also known as<00:25:55.380> a<00:25:55.410> GPU<00:25:56.540> and<00:25:57.740> probe<00:25:58.740> has<00:25:58.950> also<00:25:59.190> tunable<00:26:00.140> so 00:26:01.130 --> 00:26:01.140 align:start position:0% as a GPU and probe has also tunable so 00:26:01.140 --> 00:26:02.570 align:start position:0% as a GPU and probe has also tunable so that's<00:26:01.320> a<00:26:01.440> really<00:26:01.710> important<00:26:02.250> feature<00:26:02.370> of 00:26:02.570 --> 00:26:02.580 align:start position:0% that's a really important feature of 00:26:02.580 --> 00:26:04.400 align:start position:0% that's a really important feature of your<00:26:02.790> proof<00:26:02.970> of<00:26:03.090> work<00:26:03.240> algorithm<00:26:03.809> so<00:26:04.230> what 00:26:04.400 --> 00:26:04.410 align:start position:0% your proof of work algorithm so what 00:26:04.410 --> 00:26:07.070 align:start position:0% your proof of work algorithm so what happens<00:26:04.890> if<00:26:05.160> in<00:26:05.760> the<00:26:05.940> future<00:26:06.300> if<00:26:06.540> def<00:26:06.900> else 00:26:07.070 --> 00:26:07.080 align:start position:0% happens if in the future if def else 00:26:07.080 --> 00:26:09.260 align:start position:0% happens if in the future if def else isn't<00:26:07.410> around<00:26:07.559> to<00:26:07.679> keep<00:26:07.890> tuning<00:26:08.250> profile<00:26:08.730> for 00:26:09.260 --> 00:26:09.270 align:start position:0% isn't around to keep tuning profile for 00:26:09.270 --> 00:26:11.390 align:start position:0% isn't around to keep tuning profile for the<00:26:09.360> next<00:26:09.630> generation<00:26:10.140> of<00:26:10.350> GPUs<00:26:11.070> when<00:26:11.280> they 00:26:11.390 --> 00:26:11.400 align:start position:0% the next generation of GPUs when they 00:26:11.400 --> 00:26:12.650 align:start position:0% the next generation of GPUs when they come<00:26:11.580> out<00:26:11.700> in<00:26:11.910> six<00:26:12.179> years 00:26:12.650 --> 00:26:12.660 align:start position:0% come out in six years 00:26:12.660 --> 00:26:17.000 align:start position:0% come out in six years well<00:26:13.440> any<00:26:14.240> GPU<00:26:15.240> architect<00:26:15.960> and<00:26:16.200> their<00:26:16.410> there 00:26:17.000 --> 00:26:17.010 align:start position:0% well any GPU architect and their there 00:26:17.010 --> 00:26:19.760 align:start position:0% well any GPU architect and their there are<00:26:17.040> thousands<00:26:17.520> of<00:26:17.580> them<00:26:17.760> can<00:26:18.450> go<00:26:18.630> and<00:26:18.960> open<00:26:19.500> up 00:26:19.760 --> 00:26:19.770 align:start position:0% are thousands of them can go and open up 00:26:19.770 --> 00:26:22.700 align:start position:0% are thousands of them can go and open up the<00:26:20.340> videos<00:26:20.640> profile<00:26:21.179> or<00:26:21.390> AMD's<00:26:22.050> codec<00:26:22.470> cell 00:26:22.700 --> 00:26:22.710 align:start position:0% the videos profile or AMD's codec cell 00:26:22.710 --> 00:26:25.940 align:start position:0% the videos profile or AMD's codec cell and<00:26:22.920> simply<00:26:23.520> tune<00:26:23.760> these<00:26:23.910> parameters<00:26:24.800> to<00:26:25.800> make 00:26:25.940 --> 00:26:25.950 align:start position:0% and simply tune these parameters to make 00:26:25.950 --> 00:26:27.410 align:start position:0% and simply tune these parameters to make sure<00:26:26.190> that<00:26:26.400> the<00:26:26.520> GPU<00:26:26.880> is<00:26:27.000> completely 00:26:27.410 --> 00:26:27.420 align:start position:0% sure that the GPU is completely 00:26:27.420 --> 00:26:30.260 align:start position:0% sure that the GPU is completely saturated<00:26:27.929> and<00:26:28.520> quite<00:26:29.520> frankly<00:26:29.700> we<00:26:30.060> shouldn't 00:26:30.260 --> 00:26:30.270 align:start position:0% saturated and quite frankly we shouldn't 00:26:30.270 --> 00:26:32.090 align:start position:0% saturated and quite frankly we shouldn't be<00:26:30.450> on<00:26:30.570> proof<00:26:30.780> of<00:26:30.929> work<00:26:31.080> for<00:26:31.290> aetherium<00:26:31.620> in<00:26:31.860> six 00:26:32.090 --> 00:26:32.100 align:start position:0% be on proof of work for aetherium in six 00:26:32.100 --> 00:26:33.350 align:start position:0% be on proof of work for aetherium in six years<00:26:32.190> otherwise<00:26:32.730> I'll<00:26:32.970> be<00:26:33.090> very 00:26:33.350 --> 00:26:33.360 align:start position:0% years otherwise I'll be very 00:26:33.360 --> 00:26:35.110 align:start position:0% years otherwise I'll be very disappointed 00:26:35.110 --> 00:26:35.120 align:start position:0% disappointed 00:26:35.120 --> 00:26:38.299 align:start position:0% disappointed and<00:26:36.120> really<00:26:36.480> that's<00:26:36.929> that's<00:26:37.530> it<00:26:37.770> that's<00:26:38.100> the 00:26:38.299 --> 00:26:38.309 align:start position:0% and really that's that's it that's the 00:26:38.309 --> 00:26:40.190 align:start position:0% and really that's that's it that's the technical<00:26:38.669> walkthrough<00:26:39.240> of<00:26:39.299> Prague<00:26:39.600> pal<00:26:39.840> it 00:26:40.190 --> 00:26:40.200 align:start position:0% technical walkthrough of Prague pal it 00:26:40.200 --> 00:26:41.950 align:start position:0% technical walkthrough of Prague pal it doesn't<00:26:40.919> have<00:26:41.039> to<00:26:41.190> be<00:26:41.280> that<00:26:41.429> complex 00:26:41.950 --> 00:26:41.960 align:start position:0% doesn't have to be that complex 00:26:41.960 --> 00:26:43.100 align:start position:0% doesn't have to be that complex proof-of-work 00:26:43.100 --> 00:26:43.110 align:start position:0% proof-of-work 00:26:43.110 --> 00:26:45.649 align:start position:0% proof-of-work like<00:26:43.500> I<00:26:43.650> said<00:26:43.860> it's<00:26:44.250> just<00:26:44.460> about<00:26:44.669> matching<00:26:45.150> the 00:26:45.649 --> 00:26:45.659 align:start position:0% like I said it's just about matching the 00:26:45.659 --> 00:26:48.320 align:start position:0% like I said it's just about matching the algorithm<00:26:46.320> to<00:26:46.650> the<00:26:46.860> hardware<00:26:47.280> just<00:26:48.120> through 00:26:48.320 --> 00:26:48.330 align:start position:0% algorithm to the hardware just through 00:26:48.330 --> 00:26:53.659 align:start position:0% algorithm to the hardware just through some<00:26:48.360> minor<00:26:48.780> chains<00:26:49.580> changes<00:26:51.590> and<00:26:52.590> so<00:26:52.890> that's 00:26:53.659 --> 00:26:53.669 align:start position:0% some minor chains changes and so that's 00:26:53.669 --> 00:27:04.669 align:start position:0% some minor chains changes and so that's it<00:26:53.909> thank<00:26:54.720> you<00:26:54.840> very<00:26:54.990> much<00:27:02.690> so<00:27:03.690> I<00:27:04.080> imagine 00:27:04.669 --> 00:27:04.679 align:start position:0% it thank you very much so I imagine 00:27:04.679 --> 00:27:06.140 align:start position:0% it thank you very much so I imagine there's<00:27:05.010> going<00:27:05.070> to<00:27:05.190> be<00:27:05.250> tons<00:27:05.490> of<00:27:05.640> questions 00:27:06.140 --> 00:27:06.150 align:start position:0% there's going to be tons of questions 00:27:06.150 --> 00:27:25.190 align:start position:0% there's going to be tons of questions and<00:27:06.360> I'm<00:27:06.450> petrified<00:27:07.289> and<00:27:08.570> so<00:27:09.570> shoot<00:27:24.200> great 00:27:25.190 --> 00:27:25.200 align:start position:0% and I'm petrified and so shoot great 00:27:25.200 --> 00:27:27.260 align:start position:0% and I'm petrified and so shoot great talk<00:27:25.380> thank<00:27:25.620> you<00:27:25.740> I<00:27:26.220> want<00:27:26.909> to<00:27:26.940> play<00:27:27.090> devil's 00:27:27.260 --> 00:27:27.270 align:start position:0% talk thank you I want to play devil's 00:27:27.270 --> 00:27:30.610 align:start position:0% talk thank you I want to play devil's advocate<00:27:27.510> and<00:27:28.250> challenged<00:27:29.250> the<00:27:29.820> premise<00:27:30.179> that 00:27:30.610 --> 00:27:30.620 align:start position:0% advocate and challenged the premise that 00:27:30.620 --> 00:27:32.899 align:start position:0% advocate and challenged the premise that utilization<00:27:31.620> and<00:27:31.770> saturation<00:27:32.309> is<00:27:32.520> actually<00:27:32.880> a 00:27:32.899 --> 00:27:32.909 align:start position:0% utilization and saturation is actually a 00:27:32.909 --> 00:27:38.000 align:start position:0% utilization and saturation is actually a valid<00:27:33.780> metric<00:27:34.230> and<00:27:34.440> the<00:27:35.190> reason<00:27:35.580> is<00:27:36.020> as<00:27:37.020> a<00:27:37.289> GPU 00:27:38.000 --> 00:27:38.010 align:start position:0% valid metric and the reason is as a GPU 00:27:38.010 --> 00:27:41.600 align:start position:0% valid metric and the reason is as a GPU designer<00:27:38.929> you're<00:27:39.929> not<00:27:40.169> looking<00:27:41.010> to<00:27:41.220> solely 00:27:41.600 --> 00:27:41.610 align:start position:0% designer you're not looking to solely 00:27:41.610 --> 00:27:44.720 align:start position:0% designer you're not looking to solely optimize<00:27:42.470> energy<00:27:43.470> expenditure<00:27:44.039> per<00:27:44.190> unit<00:27:44.520> of 00:27:44.720 --> 00:27:44.730 align:start position:0% optimize energy expenditure per unit of 00:27:44.730 --> 00:27:46.789 align:start position:0% optimize energy expenditure per unit of proof<00:27:45.090> of<00:27:45.210> work<00:27:45.390> what<00:27:46.140> you're<00:27:46.289> trying<00:27:46.500> to<00:27:46.559> do 00:27:46.789 --> 00:27:46.799 align:start position:0% proof of work what you're trying to do 00:27:46.799 --> 00:27:49.430 align:start position:0% proof of work what you're trying to do is<00:27:47.070> you're<00:27:47.460> designing<00:27:47.820> for<00:27:47.909> gamers<00:27:48.440> where 00:27:49.430 --> 00:27:49.440 align:start position:0% is you're designing for gamers where 00:27:49.440 --> 00:27:52.250 align:start position:0% is you're designing for gamers where performance<00:27:50.070> is<00:27:50.340> a<00:27:50.700> key<00:27:50.970> thing<00:27:51.179> and<00:27:51.780> also<00:27:52.200> you 00:27:52.250 --> 00:27:52.260 align:start position:0% performance is a key thing and also you 00:27:52.260 --> 00:27:54.350 align:start position:0% performance is a key thing and also you have<00:27:52.500> constraints<00:27:52.950> in<00:27:53.130> terms<00:27:53.159> of<00:27:53.490> diarrhea<00:27:54.059> so 00:27:54.350 --> 00:27:54.360 align:start position:0% have constraints in terms of diarrhea so 00:27:54.360 --> 00:27:55.789 align:start position:0% have constraints in terms of diarrhea so you<00:27:54.419> want<00:27:54.630> to<00:27:54.720> optimize<00:27:55.080> for<00:27:55.440> power<00:27:55.620> so 00:27:55.789 --> 00:27:55.799 align:start position:0% you want to optimize for power so 00:27:55.799 --> 00:27:57.159 align:start position:0% you want to optimize for power so basically<00:27:56.070> what<00:27:56.220> you<00:27:56.309> do<00:27:56.490> is<00:27:56.669> that<00:27:56.700> you<00:27:56.850> will 00:27:57.159 --> 00:27:57.169 align:start position:0% basically what you do is that you will 00:27:57.169 --> 00:28:00.080 align:start position:0% basically what you do is that you will design<00:27:58.169> your<00:27:58.289> ASIC<00:27:58.650> to<00:27:58.830> have<00:27:59.039> a<00:27:59.090> reasonable 00:28:00.080 --> 00:28:00.090 align:start position:0% design your ASIC to have a reasonable 00:28:00.090 --> 00:28:03.190 align:start position:0% design your ASIC to have a reasonable trade-off<00:28:00.480> between<00:28:00.630> performance<00:28:01.400> power<00:28:02.400> and 00:28:03.190 --> 00:28:03.200 align:start position:0% trade-off between performance power and 00:28:03.200 --> 00:28:09.680 align:start position:0% trade-off between performance power and diarrhea<00:28:04.200> but<00:28:05.190> as<00:28:06.230> as<00:28:07.230> a<00:28:07.320> miner<00:28:08.390> the<00:28:09.390> only 00:28:09.680 --> 00:28:09.690 align:start position:0% diarrhea but as as a miner the only 00:28:09.690 --> 00:28:12.430 align:start position:0% diarrhea but as as a miner the only thing<00:28:09.840> you<00:28:09.960> care<00:28:10.169> about<00:28:10.200> is<00:28:10.590> power<00:28:11.419> energy 00:28:12.430 --> 00:28:12.440 align:start position:0% thing you care about is power energy 00:28:12.440 --> 00:28:16.130 align:start position:0% thing you care about is power energy spent<00:28:13.440> units<00:28:14.280> of<00:28:14.400> proof<00:28:14.580> of<00:28:14.700> work<00:28:14.850> so<00:28:15.480> you 00:28:16.130 --> 00:28:16.140 align:start position:0% spent units of proof of work so you 00:28:16.140 --> 00:28:18.620 align:start position:0% spent units of proof of work so you could<00:28:16.380> build<00:28:17.039> something<00:28:17.970> that<00:28:18.360> is 00:28:18.620 --> 00:28:18.630 align:start position:0% could build something that is 00:28:18.630 --> 00:28:21.860 align:start position:0% could build something that is functionally<00:28:19.440> equivalent<00:28:20.100> to<00:28:20.700> an<00:28:20.789> ASIC<00:28:21.120> but 00:28:21.860 --> 00:28:21.870 align:start position:0% functionally equivalent to an ASIC but 00:28:21.870 --> 00:28:24.830 align:start position:0% functionally equivalent to an ASIC but might<00:28:22.260> have<00:28:22.559> you<00:28:23.159> know<00:28:23.190> a<00:28:23.309> larger<00:28:23.850> area<00:28:24.299> or 00:28:24.830 --> 00:28:24.840 align:start position:0% might have you know a larger area or 00:28:24.840 --> 00:28:27.980 align:start position:0% might have you know a larger area or might<00:28:25.169> have<00:28:25.409> a<00:28:25.770> higher<00:28:26.100> latency<00:28:26.370> but<00:28:27.179> consume 00:28:27.980 --> 00:28:27.990 align:start position:0% might have a higher latency but consume 00:28:27.990 --> 00:28:30.500 align:start position:0% might have a higher latency but consume I'd<00:28:28.289> say<00:28:28.470> ten<00:28:28.890> times<00:28:29.039> less<00:28:29.340> energy<00:28:29.640> per<00:28:30.480> unit 00:28:30.500 --> 00:28:30.510 align:start position:0% I'd say ten times less energy per unit 00:28:30.510 --> 00:28:33.830 align:start position:0% I'd say ten times less energy per unit of<00:28:30.750> work<00:28:30.929> and<00:28:31.610> utilization<00:28:32.610> which<00:28:33.150> would<00:28:33.720> be<00:28:33.809> a 00:28:33.830 --> 00:28:33.840 align:start position:0% of work and utilization which would be a 00:28:33.840 --> 00:28:36.049 align:start position:0% of work and utilization which would be a hundred<00:28:33.990> percent<00:28:34.200> but<00:28:34.559> you<00:28:34.799> know<00:28:35.549> from<00:28:35.669> your 00:28:36.049 --> 00:28:36.059 align:start position:0% hundred percent but you know from your 00:28:36.059 --> 00:28:38.630 align:start position:0% hundred percent but you know from your perspective<00:28:36.179> that<00:28:36.780> it's<00:28:37.230> not<00:28:37.380> a<00:28:37.980> good<00:28:38.460> metric 00:28:38.630 --> 00:28:38.640 align:start position:0% perspective that it's not a good metric 00:28:38.640 --> 00:28:40.730 align:start position:0% perspective that it's not a good metric so<00:28:39.360> there<00:28:39.809> are<00:28:39.929> three<00:28:40.140> things<00:28:40.320> to<00:28:40.559> address 00:28:40.730 --> 00:28:40.740 align:start position:0% so there are three things to address 00:28:40.740 --> 00:28:44.450 align:start position:0% so there are three things to address there<00:28:41.809> one<00:28:42.809> is<00:28:42.929> a<00:28:43.070> philosophical<00:28:44.070> thing<00:28:44.130> where 00:28:44.450 --> 00:28:44.460 align:start position:0% there one is a philosophical thing where 00:28:44.460 --> 00:28:45.980 align:start position:0% there one is a philosophical thing where we<00:28:44.549> talk<00:28:44.700> about<00:28:44.880> what<00:28:45.299> proof<00:28:45.630> of<00:28:45.780> work 00:28:45.980 --> 00:28:45.990 align:start position:0% we talk about what proof of work 00:28:45.990 --> 00:28:47.340 align:start position:0% we talk about what proof of work actually<00:28:46.559> meant<00:28:46.830> when<00:28:46.980> it<00:28:47.070> was<00:28:47.159> a 00:28:47.340 --> 00:28:47.350 align:start position:0% actually meant when it was a 00:28:47.350 --> 00:28:50.760 align:start position:0% actually meant when it was a created<00:28:48.010> you<00:28:48.610> know<00:28:48.730> proof-of-work<00:28:49.650> kind<00:28:50.650> of 00:28:50.760 --> 00:28:50.770 align:start position:0% created you know proof-of-work kind of 00:28:50.770 --> 00:28:53.820 align:start position:0% created you know proof-of-work kind of means<00:28:50.950> work<00:28:51.220> AKA<00:28:51.850> energy<00:28:52.419> expended<00:28:52.990> but 00:28:53.820 --> 00:28:53.830 align:start position:0% means work AKA energy expended but 00:28:53.830 --> 00:28:55.260 align:start position:0% means work AKA energy expended but that's<00:28:53.980> really<00:28:54.190> a<00:28:54.280> philosophical<00:28:54.549> debate 00:28:55.260 --> 00:28:55.270 align:start position:0% that's really a philosophical debate 00:28:55.270 --> 00:28:57.930 align:start position:0% that's really a philosophical debate from<00:28:56.169> a<00:28:56.260> minders<00:28:56.650> stand<00:28:56.980> for<00:28:57.520> miners 00:28:57.930 --> 00:28:57.940 align:start position:0% from a minders stand for miners 00:28:57.940 --> 00:29:00.960 align:start position:0% from a minders stand for miners standpoint<00:28:59.100> miners<00:29:00.100> really<00:29:00.520> care<00:29:00.789> about 00:29:00.960 --> 00:29:00.970 align:start position:0% standpoint miners really care about 00:29:00.970 --> 00:29:04.110 align:start position:0% standpoint miners really care about balance<00:29:01.720> now<00:29:02.440> prog<00:29:02.919> pal<00:29:03.220> it<00:29:03.549> does<00:29:03.760> consume 00:29:04.110 --> 00:29:04.120 align:start position:0% balance now prog pal it does consume 00:29:04.120 --> 00:29:07.080 align:start position:0% balance now prog pal it does consume more<00:29:04.390> more<00:29:04.870> power<00:29:05.080> simply<00:29:05.770> because<00:29:06.130> it<00:29:06.280> uses 00:29:07.080 --> 00:29:07.090 align:start position:0% more more power simply because it uses 00:29:07.090 --> 00:29:09.810 align:start position:0% more more power simply because it uses the<00:29:07.480> core<00:29:07.870> which<00:29:08.110> was<00:29:08.320> not<00:29:08.500> used<00:29:08.679> before<00:29:08.890> but 00:29:09.810 --> 00:29:09.820 align:start position:0% the core which was not used before but 00:29:09.820 --> 00:29:12.570 align:start position:0% the core which was not used before but the<00:29:10.240> important<00:29:10.720> thing<00:29:10.870> with<00:29:10.990> that<00:29:11.169> is<00:29:11.580> that 00:29:12.570 --> 00:29:12.580 align:start position:0% the important thing with that is that 00:29:12.580 --> 00:29:16.080 align:start position:0% the important thing with that is that you<00:29:13.330> take<00:29:13.539> an<00:29:13.720> ASIC<00:29:14.070> that<00:29:15.070> also<00:29:15.340> has<00:29:15.669> the<00:29:15.880> core 00:29:16.080 --> 00:29:16.090 align:start position:0% you take an ASIC that also has the core 00:29:16.090 --> 00:29:19.049 align:start position:0% you take an ASIC that also has the core also<00:29:16.419> has<00:29:16.809> the<00:29:16.960> memory<00:29:17.140> take<00:29:17.770> a<00:29:17.799> GPU<00:29:18.190> same<00:29:18.820> it 00:29:19.049 --> 00:29:19.059 align:start position:0% also has the memory take a GPU same it 00:29:19.059 --> 00:29:21.960 align:start position:0% also has the memory take a GPU same it will<00:29:19.270> match<00:29:20.130> your<00:29:21.130> other<00:29:21.340> point<00:29:21.669> was<00:29:21.909> that 00:29:21.960 --> 00:29:21.970 align:start position:0% will match your other point was that 00:29:21.970 --> 00:29:25.710 align:start position:0% will match your other point was that miners<00:29:22.720> only<00:29:23.200> care<00:29:23.500> about<00:29:23.530> power<00:29:24.070> but<00:29:24.640> I'd<00:29:25.090> I'd 00:29:25.710 --> 00:29:25.720 align:start position:0% miners only care about power but I'd I'd 00:29:25.720 --> 00:29:30.270 align:start position:0% miners only care about power but I'd I'd really<00:29:26.140> contest<00:29:26.530> that<00:29:27.900> miners<00:29:28.919> really<00:29:29.919> really 00:29:30.270 --> 00:29:30.280 align:start position:0% really contest that miners really really 00:29:30.280 --> 00:29:33.539 align:start position:0% really contest that miners really really just<00:29:30.460> care<00:29:30.730> about<00:29:30.760> fairness<00:29:32.100> that's<00:29:33.100> that's 00:29:33.539 --> 00:29:33.549 align:start position:0% just care about fairness that's that's 00:29:33.549 --> 00:29:35.700 align:start position:0% just care about fairness that's that's the<00:29:33.730> crux<00:29:33.940> of<00:29:34.120> it<00:29:34.210> we<00:29:34.539> care<00:29:34.780> about<00:29:34.960> being<00:29:35.470> able 00:29:35.700 --> 00:29:35.710 align:start position:0% the crux of it we care about being able 00:29:35.710 --> 00:29:38.580 align:start position:0% the crux of it we care about being able to<00:29:36.159> all<00:29:37.000> be<00:29:37.330> on<00:29:37.450> the<00:29:37.480> same<00:29:37.840> playing<00:29:38.110> field<00:29:38.230> in 00:29:38.580 --> 00:29:38.590 align:start position:0% to all be on the same playing field in 00:29:38.590 --> 00:29:41.010 align:start position:0% to all be on the same playing field in the<00:29:39.549> same<00:29:39.760> level<00:29:39.970> field<00:29:40.330> and<00:29:40.630> then<00:29:40.929> you 00:29:41.010 --> 00:29:41.020 align:start position:0% the same level field and then you 00:29:41.020 --> 00:29:43.110 align:start position:0% the same level field and then you touched<00:29:41.289> on<00:29:41.500> the<00:29:41.620> fact<00:29:41.650> that<00:29:41.950> GPUs<00:29:42.549> are<00:29:42.730> just 00:29:43.110 --> 00:29:43.120 align:start position:0% touched on the fact that GPUs are just 00:29:43.120 --> 00:29:44.669 align:start position:0% touched on the fact that GPUs are just naturally<00:29:43.510> tuned<00:29:43.720> for<00:29:43.929> video<00:29:44.169> games<00:29:44.440> that's 00:29:44.669 --> 00:29:44.679 align:start position:0% naturally tuned for video games that's 00:29:44.679 --> 00:29:47.520 align:start position:0% naturally tuned for video games that's not<00:29:44.890> true<00:29:45.100> at<00:29:45.190> all<00:29:45.340> see<00:29:46.059> GPUs<00:29:46.480> are<00:29:46.659> tuned<00:29:47.200> for 00:29:47.520 --> 00:29:47.530 align:start position:0% not true at all see GPUs are tuned for 00:29:47.530 --> 00:29:52.169 align:start position:0% not true at all see GPUs are tuned for maths<00:29:48.419> geforce<00:29:49.419> cards<00:29:50.080> and<00:29:50.350> AMD's<00:29:51.250> gaming 00:29:52.169 --> 00:29:52.179 align:start position:0% maths geforce cards and AMD's gaming 00:29:52.179 --> 00:29:55.560 align:start position:0% maths geforce cards and AMD's gaming line<00:29:52.360> sure<00:29:52.809> it's<00:29:52.960> tuned<00:29:53.289> for<00:29:53.559> FPS<00:29:54.340> but<00:29:55.179> by 00:29:55.560 --> 00:29:55.570 align:start position:0% line sure it's tuned for FPS but by 00:29:55.570 --> 00:29:57.210 align:start position:0% line sure it's tuned for FPS but by modifying<00:29:55.990> the<00:29:56.169> firmware<00:29:56.559> you<00:29:56.710> can<00:29:56.830> also<00:29:56.950> just 00:29:57.210 --> 00:29:57.220 align:start position:0% modifying the firmware you can also just 00:29:57.220 --> 00:29:59.039 align:start position:0% modifying the firmware you can also just tune<00:29:57.460> it<00:29:57.549> for<00:29:57.580> math<00:29:57.880> that's<00:29:58.570> the<00:29:58.720> great<00:29:59.020> thing 00:29:59.039 --> 00:29:59.049 align:start position:0% tune it for math that's the great thing 00:29:59.049 --> 00:30:01.140 align:start position:0% tune it for math that's the great thing about<00:29:59.320> GPUs<00:29:59.919> they're<00:30:00.130> tunable<00:30:01.030> and 00:30:01.140 --> 00:30:01.150 align:start position:0% about GPUs they're tunable and 00:30:01.150 --> 00:30:04.680 align:start position:0% about GPUs they're tunable and programmable<00:30:01.690> and<00:30:01.929> customizable<00:30:03.539> what<00:30:04.539> was 00:30:04.680 --> 00:30:04.690 align:start position:0% programmable and customizable what was 00:30:04.690 --> 00:30:06.360 align:start position:0% programmable and customizable what was what<00:30:05.140> was<00:30:05.320> the<00:30:05.350> other<00:30:05.590> parts<00:30:06.010> of<00:30:06.309> your 00:30:06.360 --> 00:30:06.370 align:start position:0% what was the other parts of your 00:30:06.370 --> 00:30:15.700 align:start position:0% what was the other parts of your question 00:30:15.700 --> 00:30:15.710 align:start position:0% 00:30:15.710 --> 00:30:24.169 align:start position:0% you<00:30:20.480> can<00:30:21.480> try<00:30:21.720> and<00:30:21.870> minimize<00:30:22.650> latency<00:30:23.490> you<00:30:24.030> can 00:30:24.169 --> 00:30:24.179 align:start position:0% you can try and minimize latency you can 00:30:24.179 --> 00:30:26.390 align:start position:0% you can try and minimize latency you can try<00:30:24.360> and<00:30:24.480> minimize<00:30:24.870> area<00:30:25.590> or<00:30:25.799> you<00:30:25.950> can<00:30:26.130> try<00:30:26.340> and 00:30:26.390 --> 00:30:26.400 align:start position:0% try and minimize area or you can try and 00:30:26.400 --> 00:30:29.150 align:start position:0% try and minimize area or you can try and minimize<00:30:26.850> power<00:30:27.650> you<00:30:28.650> could<00:30:28.799> find<00:30:29.039> a 00:30:29.150 --> 00:30:29.160 align:start position:0% minimize power you could find a 00:30:29.160 --> 00:30:31.610 align:start position:0% minimize power you could find a reasonable<00:30:29.610> trade-off<00:30:30.120> if<00:30:30.809> I<00:30:31.110> was<00:30:31.380> to<00:30:31.500> go 00:30:31.610 --> 00:30:31.620 align:start position:0% reasonable trade-off if I was to go 00:30:31.620 --> 00:30:34.880 align:start position:0% reasonable trade-off if I was to go ahead<00:30:32.010> and<00:30:32.429> build<00:30:32.640> an<00:30:32.700> ASIC<00:30:33.059> for<00:30:33.679> pro<00:30:34.679> power 00:30:34.880 --> 00:30:34.890 align:start position:0% ahead and build an ASIC for pro power 00:30:34.890 --> 00:30:40.419 align:start position:0% ahead and build an ASIC for pro power what<00:30:35.549> I<00:30:35.610> would<00:30:35.789> do<00:30:36.000> is<00:30:36.330> I<00:30:36.840> would<00:30:38.600> make<00:30:39.600> it 00:30:40.419 --> 00:30:40.429 align:start position:0% what I would do is I would make it 00:30:40.429 --> 00:30:43.340 align:start position:0% what I would do is I would make it energy<00:30:41.429> efficient<00:30:41.880> first<00:30:42.150> so<00:30:42.720> I<00:30:42.750> would<00:30:43.020> choose 00:30:43.340 --> 00:30:43.350 align:start position:0% energy efficient first so I would choose 00:30:43.350 --> 00:30:47.029 align:start position:0% energy efficient first so I would choose my<00:30:43.620> cells<00:30:44.280> extremely<00:30:44.820> carefully<00:30:45.270> so<00:30:45.960> that<00:30:46.039> you 00:30:47.029 --> 00:30:47.039 align:start position:0% my cells extremely carefully so that you 00:30:47.039 --> 00:30:49.430 align:start position:0% my cells extremely carefully so that you know<00:30:47.160> I<00:30:47.190> don't<00:30:47.309> care<00:30:47.700> if<00:30:47.880> the<00:30:48.150> 10x<00:30:48.780> the<00:30:48.960> latency 00:30:49.430 --> 00:30:49.440 align:start position:0% know I don't care if the 10x the latency 00:30:49.440 --> 00:30:53.600 align:start position:0% know I don't care if the 10x the latency or<00:30:49.710> if<00:30:50.100> the<00:30:50.309> 3x<00:30:50.940> the<00:30:51.510> die<00:30:51.690> area<00:30:52.130> all<00:30:53.130> I<00:30:53.190> care 00:30:53.600 --> 00:30:53.610 align:start position:0% or if the 3x the die area all I care 00:30:53.610 --> 00:30:56.419 align:start position:0% or if the 3x the die area all I care about<00:30:53.640> because<00:30:54.179> the<00:30:54.870> revenue<00:30:55.650> as<00:30:55.770> a<00:30:55.799> minor<00:30:56.100> is 00:30:56.419 --> 00:30:56.429 align:start position:0% about because the revenue as a minor is 00:30:56.429 --> 00:30:59.980 align:start position:0% about because the revenue as a minor is basically<00:30:57.049> going<00:30:58.049> to<00:30:58.140> be<00:30:58.230> proportional<00:30:58.650> to 00:30:59.980 --> 00:30:59.990 align:start position:0% basically going to be proportional to 00:30:59.990 --> 00:31:05.539 align:start position:0% basically going to be proportional to the<00:31:01.130> the<00:31:02.130> amount<00:31:03.470> of<00:31:04.470> work<00:31:04.890> that<00:31:05.159> you<00:31:05.220> can<00:31:05.340> get 00:31:05.539 --> 00:31:05.549 align:start position:0% the the amount of work that you can get 00:31:05.549 --> 00:31:08.960 align:start position:0% the the amount of work that you can get done<00:31:05.820> drool<00:31:06.510> can<00:31:06.870> burn<00:31:07.340> only<00:31:08.340> when<00:31:08.820> it's 00:31:08.960 --> 00:31:08.970 align:start position:0% done drool can burn only when it's 00:31:08.970 --> 00:31:11.510 align:start position:0% done drool can burn only when it's compared<00:31:09.419> to<00:31:09.630> other<00:31:09.809> miners<00:31:10.380> of<00:31:11.309> course 00:31:11.510 --> 00:31:11.520 align:start position:0% compared to other miners of course 00:31:11.520 --> 00:31:13.460 align:start position:0% compared to other miners of course that's<00:31:11.970> that's<00:31:12.330> the<00:31:12.539> difference<00:31:12.929> yes<00:31:13.140> so<00:31:13.350> the 00:31:13.460 --> 00:31:13.470 align:start position:0% that's that's the difference yes so the 00:31:13.470 --> 00:31:16.789 align:start position:0% that's that's the difference yes so the other<00:31:13.620> miners<00:31:14.010> will<00:31:14.220> use<00:31:14.750> GPUs<00:31:15.750> and<00:31:15.990> I<00:31:16.590> will 00:31:16.789 --> 00:31:16.799 align:start position:0% other miners will use GPUs and I will 00:31:16.799 --> 00:31:19.220 align:start position:0% other miners will use GPUs and I will use<00:31:17.130> something<00:31:17.669> that<00:31:17.909> is<00:31:18.230> functionally 00:31:19.220 --> 00:31:19.230 align:start position:0% use something that is functionally 00:31:19.230 --> 00:31:23.720 align:start position:0% use something that is functionally equivalent<00:31:19.590> to<00:31:20.220> GPU<00:31:20.610> but<00:31:21.360> is<00:31:22.460> potentially<00:31:23.460> has 00:31:23.720 --> 00:31:23.730 align:start position:0% equivalent to GPU but is potentially has 00:31:23.730 --> 00:31:29.000 align:start position:0% equivalent to GPU but is potentially has a<00:31:23.760> much<00:31:24.120> larger<00:31:24.360> diarrhea<00:31:25.080> and<00:31:25.740> runs<00:31:27.830> but<00:31:28.830> when 00:31:29.000 --> 00:31:29.010 align:start position:0% a much larger diarrhea and runs but when 00:31:29.010 --> 00:31:31.070 align:start position:0% a much larger diarrhea and runs but when you<00:31:29.190> when<00:31:29.490> you<00:31:29.640> go<00:31:29.789> and<00:31:29.909> build<00:31:30.059> a<00:31:30.210> propel<00:31:30.630> ASIC 00:31:31.070 --> 00:31:31.080 align:start position:0% you when you go and build a propel ASIC 00:31:31.080 --> 00:31:32.570 align:start position:0% you when you go and build a propel ASIC you're<00:31:31.260> only<00:31:31.440> going<00:31:31.710> to<00:31:31.830> be<00:31:31.950> able<00:31:32.130> to<00:31:32.250> shave 00:31:32.570 --> 00:31:32.580 align:start position:0% you're only going to be able to shave 00:31:32.580 --> 00:31:34.669 align:start position:0% you're only going to be able to shave off<00:31:32.820> I<00:31:33.179> get<00:31:33.570> your<00:31:33.690> question<00:31:34.020> now<00:31:34.140> you're 00:31:34.669 --> 00:31:34.679 align:start position:0% off I get your question now you're 00:31:34.679 --> 00:31:35.870 align:start position:0% off I get your question now you're talking<00:31:35.010> about<00:31:35.070> the<00:31:35.220> fact<00:31:35.490> that<00:31:35.549> you're<00:31:35.730> going 00:31:35.870 --> 00:31:35.880 align:start position:0% talking about the fact that you're going 00:31:35.880 --> 00:31:37.880 align:start position:0% talking about the fact that you're going to<00:31:35.909> shave<00:31:36.179> off<00:31:36.360> although<00:31:36.750> you<00:31:37.320> know<00:31:37.409> the<00:31:37.559> GPU 00:31:37.880 --> 00:31:37.890 align:start position:0% to shave off although you know the GPU 00:31:37.890 --> 00:31:39.529 align:start position:0% to shave off although you know the GPU pipelines<00:31:38.400> the<00:31:38.580> extra<00:31:38.940> Katia's<00:31:39.270> all<00:31:39.419> the 00:31:39.529 --> 00:31:39.539 align:start position:0% pipelines the extra Katia's all the 00:31:39.539 --> 00:31:41.419 align:start position:0% pipelines the extra Katia's all the things<00:31:39.780> we've<00:31:39.990> left<00:31:40.260> to<00:31:40.799> optimize<00:31:41.220> it<00:31:41.400> for 00:31:41.419 --> 00:31:41.429 align:start position:0% things we've left to optimize it for 00:31:41.429 --> 00:31:42.080 align:start position:0% things we've left to optimize it for power 00:31:42.080 --> 00:31:42.090 align:start position:0% power 00:31:42.090 --> 00:31:44.180 align:start position:0% power it's<00:31:42.419> not<00:31:42.570> about<00:31:42.690> shaving<00:31:43.200> of<00:31:43.530> things<00:31:43.890> is 00:31:44.180 --> 00:31:44.190 align:start position:0% it's not about shaving of things is 00:31:44.190 --> 00:31:48.020 align:start position:0% it's not about shaving of things is about<00:31:45.230> fundamentally<00:31:46.610> designing<00:31:47.610> the 00:31:48.020 --> 00:31:48.030 align:start position:0% about fundamentally designing the 00:31:48.030 --> 00:31:51.470 align:start position:0% about fundamentally designing the circuit<00:31:49.070> optimized<00:31:50.070> for<00:31:50.460> power<00:31:50.820> as<00:31:51.210> opposed 00:31:51.470 --> 00:31:51.480 align:start position:0% circuit optimized for power as opposed 00:31:51.480 --> 00:31:53.419 align:start position:0% circuit optimized for power as opposed to<00:31:51.539> optimizing<00:31:52.080> you<00:31:52.350> can't<00:31:52.679> really<00:31:52.950> optimize 00:31:53.419 --> 00:31:53.429 align:start position:0% to optimizing you can't really optimize 00:31:53.429 --> 00:31:55.399 align:start position:0% to optimizing you can't really optimize for<00:31:53.730> power<00:31:53.880> when<00:31:54.240> you<00:31:54.360> just<00:31:54.600> have<00:31:54.809> a<00:31:54.840> bunch<00:31:55.200> of 00:31:55.399 --> 00:31:55.409 align:start position:0% for power when you just have a bunch of 00:31:55.409 --> 00:31:57.830 align:start position:0% for power when you just have a bunch of random<00:31:56.190> math<00:31:56.940> churning<00:31:57.419> through<00:31:57.659> there 00:31:57.830 --> 00:31:57.840 align:start position:0% random math churning through there 00:31:57.840 --> 00:31:59.779 align:start position:0% random math churning through there that's<00:31:58.440> that's<00:31:58.770> going<00:31:59.010> to<00:31:59.070> consume<00:31:59.370> a<00:31:59.549> lot<00:31:59.700> of 00:31:59.779 --> 00:31:59.789 align:start position:0% that's that's going to consume a lot of 00:31:59.789 --> 00:32:01.669 align:start position:0% that's that's going to consume a lot of power<00:32:00.030> and<00:32:00.240> no<00:32:00.360> matter<00:32:00.539> what<00:32:00.780> the<00:32:01.350> only<00:32:01.559> way 00:32:01.669 --> 00:32:01.679 align:start position:0% power and no matter what the only way 00:32:01.679 --> 00:32:03.580 align:start position:0% power and no matter what the only way you<00:32:01.740> can<00:32:01.980> get<00:32:02.100> the<00:32:02.250> power<00:32:02.640> savings<00:32:03.240> in<00:32:03.480> that 00:32:03.580 --> 00:32:03.590 align:start position:0% you can get the power savings in that 00:32:03.590 --> 00:32:06.950 align:start position:0% you can get the power savings in that circuit<00:32:04.610> would<00:32:05.610> be<00:32:05.820> to<00:32:06.150> probably<00:32:06.330> shave<00:32:06.750> off<00:32:06.929> a 00:32:06.950 --> 00:32:06.960 align:start position:0% circuit would be to probably shave off a 00:32:06.960 --> 00:32:10.250 align:start position:0% circuit would be to probably shave off a few<00:32:07.260> transistors<00:32:07.740> let<00:32:08.220> me<00:32:08.309> give<00:32:08.460> you<00:32:09.260> let's 00:32:10.250 --> 00:32:10.260 align:start position:0% few transistors let me give you let's 00:32:10.260 --> 00:32:14.240 align:start position:0% few transistors let me give you let's say<00:32:10.530> you<00:32:11.220> want<00:32:11.370> to<00:32:11.460> build<00:32:11.669> a<00:32:13.159> 256<00:32:14.159> bit 00:32:14.240 --> 00:32:14.250 align:start position:0% say you want to build a 256 bit 00:32:14.250 --> 00:32:16.610 align:start position:0% say you want to build a 256 bit multiplier<00:32:15.000> mm-hmm<00:32:15.659> you<00:32:16.140> have<00:32:16.289> two<00:32:16.440> options 00:32:16.610 --> 00:32:16.620 align:start position:0% multiplier mm-hmm you have two options 00:32:16.620 --> 00:32:19.730 align:start position:0% multiplier mm-hmm you have two options one<00:32:17.130> is<00:32:17.490> that<00:32:17.909> you<00:32:18.030> can<00:32:18.240> build<00:32:18.480> a<00:32:18.740> massively 00:32:19.730 --> 00:32:19.740 align:start position:0% one is that you can build a massively 00:32:19.740 --> 00:32:22.159 align:start position:0% one is that you can build a massively parallel<00:32:20.419> multiplier<00:32:21.419> where<00:32:21.750> basically 00:32:22.159 --> 00:32:22.169 align:start position:0% parallel multiplier where basically 00:32:22.169 --> 00:32:25.700 align:start position:0% parallel multiplier where basically you'll<00:32:22.470> have<00:32:22.700> 256<00:32:23.700> by<00:32:24.210> 256<00:32:24.480> array<00:32:25.350> and<00:32:25.650> then 00:32:25.700 --> 00:32:25.710 align:start position:0% you'll have 256 by 256 array and then 00:32:25.710 --> 00:32:26.510 align:start position:0% you'll have 256 by 256 array and then you'll<00:32:25.860> have<00:32:26.010> a<00:32:26.070> massive 00:32:26.510 --> 00:32:26.520 align:start position:0% you'll have a massive 00:32:26.520 --> 00:32:28.220 align:start position:0% you'll have a massive reduction<00:32:26.910> trees<00:32:27.180> with<00:32:27.420> lots<00:32:27.540> of<00:32:27.690> compressors 00:32:28.220 --> 00:32:28.230 align:start position:0% reduction trees with lots of compressors 00:32:28.230 --> 00:32:30.440 align:start position:0% reduction trees with lots of compressors and<00:32:28.440> that's<00:32:29.100> going<00:32:29.280> to<00:32:29.340> use<00:32:29.610> a<00:32:29.640> lot<00:32:29.910> of<00:32:30.000> dynamic 00:32:30.440 --> 00:32:30.450 align:start position:0% and that's going to use a lot of dynamic 00:32:30.450 --> 00:32:31.340 align:start position:0% and that's going to use a lot of dynamic power<00:32:30.660> mm-hmm 00:32:31.340 --> 00:32:31.350 align:start position:0% power mm-hmm 00:32:31.350 --> 00:32:33.470 align:start position:0% power mm-hmm the<00:32:31.830> other<00:32:31.980> option<00:32:32.340> is<00:32:32.370> to<00:32:32.610> have<00:32:32.760> a<00:32:32.820> much 00:32:33.470 --> 00:32:33.480 align:start position:0% the other option is to have a much 00:32:33.480 --> 00:32:35.600 align:start position:0% the other option is to have a much smaller<00:32:33.630> circuit<00:32:34.260> which<00:32:34.980> does<00:32:35.130> less<00:32:35.370> things 00:32:35.600 --> 00:32:35.610 align:start position:0% smaller circuit which does less things 00:32:35.610 --> 00:32:38.300 align:start position:0% smaller circuit which does less things in<00:32:35.730> parallel<00:32:36.000> like<00:32:36.270> much<00:32:36.960> more<00:32:37.290> much<00:32:38.190> more 00:32:38.300 --> 00:32:38.310 align:start position:0% in parallel like much more much more 00:32:38.310 --> 00:32:40.880 align:start position:0% in parallel like much more much more sequential<00:32:38.670> work<00:32:38.970> and<00:32:39.870> that<00:32:40.140> is<00:32:40.470> going<00:32:40.800> to 00:32:40.880 --> 00:32:40.890 align:start position:0% sequential work and that is going to 00:32:40.890 --> 00:32:45.260 align:start position:0% sequential work and that is going to consume<00:32:41.280> less<00:32:41.670> energy<00:32:44.270> 256-bit 00:32:45.260 --> 00:32:45.270 align:start position:0% consume less energy 256-bit 00:32:45.270 --> 00:32:47.720 align:start position:0% consume less energy 256-bit multiplication<00:32:46.230> now<00:32:46.410> it's<00:32:47.040> possible<00:32:47.310> and 00:32:47.720 --> 00:32:47.730 align:start position:0% multiplication now it's possible and 00:32:47.730 --> 00:32:51.160 align:start position:0% multiplication now it's possible and it's<00:32:48.420> likely<00:32:48.630> that<00:32:49.130> GPU<00:32:50.130> designers<00:32:50.640> did<00:32:50.850> not 00:32:51.160 --> 00:32:51.170 align:start position:0% it's likely that GPU designers did not 00:32:51.170 --> 00:32:55.190 align:start position:0% it's likely that GPU designers did not build<00:32:52.170> their<00:32:52.470> circuits<00:32:52.920> to<00:32:53.190> minimize<00:32:53.760> for<00:32:54.200> for 00:32:55.190 --> 00:32:55.200 align:start position:0% build their circuits to minimize for for 00:32:55.200 --> 00:32:59.060 align:start position:0% build their circuits to minimize for for low<00:32:55.500> power<00:32:56.750> for<00:32:57.750> low<00:32:57.960> energy<00:32:58.170> expenditure<00:32:58.830> per 00:32:59.060 --> 00:32:59.070 align:start position:0% low power for low energy expenditure per 00:32:59.070 --> 00:33:01.370 align:start position:0% low power for low energy expenditure per unit<00:32:59.400> of<00:32:59.520> work<00:32:59.700> I<00:33:00.030> wouldn't<00:33:00.720> say<00:33:00.900> that 00:33:01.370 --> 00:33:01.380 align:start position:0% unit of work I wouldn't say that 00:33:01.380 --> 00:33:04.520 align:start position:0% unit of work I wouldn't say that I'd<00:33:01.620> go<00:33:01.890> and<00:33:02.070> look<00:33:02.250> at<00:33:02.400> the<00:33:03.090> GPUs<00:33:03.540> design<00:33:03.960> again 00:33:04.520 --> 00:33:04.530 align:start position:0% I'd go and look at the GPUs design again 00:33:04.530 --> 00:33:06.440 align:start position:0% I'd go and look at the GPUs design again the<00:33:04.800> most<00:33:05.040> the<00:33:05.280> ones<00:33:05.460> we<00:33:05.670> target<00:33:06.030> here<00:33:06.180> are<00:33:06.300> G 00:33:06.440 --> 00:33:06.450 align:start position:0% the most the ones we target here are G 00:33:06.450 --> 00:33:08.330 align:start position:0% the most the ones we target here are G Falls<00:33:06.690> but<00:33:07.230> if<00:33:07.290> you<00:33:07.470> actually<00:33:07.620> go<00:33:07.800> and<00:33:07.920> look<00:33:08.100> at 00:33:08.330 --> 00:33:08.340 align:start position:0% Falls but if you actually go and look at 00:33:08.340 --> 00:33:10.910 align:start position:0% Falls but if you actually go and look at AMD<00:33:08.610> or<00:33:08.880> nvidia<00:33:09.270> gpus<00:33:09.870> that<00:33:10.140> do<00:33:10.320> target<00:33:10.770> low 00:33:10.910 --> 00:33:10.920 align:start position:0% AMD or nvidia gpus that do target low 00:33:10.920 --> 00:33:12.950 align:start position:0% AMD or nvidia gpus that do target low power<00:33:11.190> for<00:33:11.850> every<00:33:12.090> energy<00:33:12.420> of<00:33:12.510> work<00:33:12.720> it's 00:33:12.950 --> 00:33:12.960 align:start position:0% power for every energy of work it's 00:33:12.960 --> 00:33:15.650 align:start position:0% power for every energy of work it's their<00:33:13.140> new<00:33:13.320> tesla<00:33:13.860> lines<00:33:14.330> specifically<00:33:15.330> for 00:33:15.650 --> 00:33:15.660 align:start position:0% their new tesla lines specifically for 00:33:15.660 --> 00:33:19.390 align:start position:0% their new tesla lines specifically for g6<00:33:16.170> that<00:33:16.410> consume<00:33:16.770> about<00:33:17.060> 55<00:33:18.060> watts<00:33:18.420> per<00:33:18.750> card 00:33:19.390 --> 00:33:19.400 align:start position:0% g6 that consume about 55 watts per card 00:33:19.400 --> 00:33:23.650 align:start position:0% g6 that consume about 55 watts per card for<00:33:20.400> about<00:33:20.640> I<00:33:21.210> think<00:33:21.720> it's<00:33:21.930> 9.6<00:33:22.800> teraflops 00:33:23.650 --> 00:33:23.660 align:start position:0% for about I think it's 9.6 teraflops 00:33:23.660 --> 00:33:26.600 align:start position:0% for about I think it's 9.6 teraflops that's<00:33:24.660> a<00:33:24.810> great<00:33:25.020> example<00:33:25.290> of<00:33:25.830> optimizing<00:33:26.310> for 00:33:26.600 --> 00:33:26.610 align:start position:0% that's a great example of optimizing for 00:33:26.610 --> 00:33:28.880 align:start position:0% that's a great example of optimizing for power<00:33:26.790> it's<00:33:27.390> just<00:33:27.630> about<00:33:27.810> which<00:33:28.020> line<00:33:28.290> of<00:33:28.440> GPUs 00:33:28.880 --> 00:33:28.890 align:start position:0% power it's just about which line of GPUs 00:33:28.890 --> 00:33:32.540 align:start position:0% power it's just about which line of GPUs you're<00:33:29.070> going<00:33:29.340> to<00:33:29.430> target<00:33:30.440> so<00:33:31.440> they<00:33:31.560> do<00:33:31.620> build 00:33:32.540 --> 00:33:32.550 align:start position:0% you're going to target so they do build 00:33:32.550 --> 00:33:39.980 align:start position:0% you're going to target so they do build some<00:33:32.760> things<00:33:33.000> like<00:33:33.150> that<00:33:37.460> high<00:33:38.660> so<00:33:39.660> there<00:33:39.840> is<00:33:39.960> a 00:33:39.980 --> 00:33:39.990 align:start position:0% some things like that high so there is a 00:33:39.990 --> 00:33:42.620 align:start position:0% some things like that high so there is a speculation<00:33:40.550> that<00:33:41.550> this<00:33:41.760> algorithm<00:33:42.210> was 00:33:42.620 --> 00:33:42.630 align:start position:0% speculation that this algorithm was 00:33:42.630 --> 00:33:47.570 align:start position:0% speculation that this algorithm was developed<00:33:42.690> by<00:33:43.230> AMD<00:33:43.470> and<00:33:43.980> in<00:33:44.100> media<00:33:46.430> in<00:33:47.430> order 00:33:47.570 --> 00:33:47.580 align:start position:0% developed by AMD and in media in order 00:33:47.580 --> 00:33:51.530 align:start position:0% developed by AMD and in media in order to<00:33:47.730> protect<00:33:48.030> GPU<00:33:48.900> money<00:33:49.140> and<00:33:49.430> interests<00:33:50.540> so 00:33:51.530 --> 00:33:51.540 align:start position:0% to protect GPU money and interests so 00:33:51.540 --> 00:33:54.050 align:start position:0% to protect GPU money and interests so and<00:33:51.870> you<00:33:52.260> did<00:33:52.440> obfuscated<00:33:53.220> people<00:33:53.700> who<00:33:53.850> worked 00:33:54.050 --> 00:33:54.060 align:start position:0% and you did obfuscated people who worked 00:33:54.060 --> 00:33:55.280 align:start position:0% and you did obfuscated people who worked who<00:33:54.510> is<00:33:54.570> behind<00:33:54.900> it 00:33:55.280 --> 00:33:55.290 align:start position:0% who is behind it 00:33:55.290 --> 00:33:58.580 align:start position:0% who is behind it but<00:33:55.410> if<00:33:56.310> then<00:33:56.610> else<00:33:56.850> why<00:33:57.750> does<00:33:57.960> it<00:33:58.110> matter<00:33:58.230> who 00:33:58.580 --> 00:33:58.590 align:start position:0% but if then else why does it matter who 00:33:58.590 --> 00:34:01.900 align:start position:0% but if then else why does it matter who created<00:33:59.010> it<00:33:59.220> did<00:33:59.640> you<00:33:59.820> so<00:34:00.360> two<00:34:01.020> parts<00:34:01.200> to<00:34:01.440> that 00:34:01.900 --> 00:34:01.910 align:start position:0% created it did you so two parts to that 00:34:01.910 --> 00:34:05.030 align:start position:0% created it did you so two parts to that first<00:34:02.910> the<00:34:03.540> amount<00:34:03.840> of<00:34:04.020> work<00:34:04.350> and<00:34:04.620> research 00:34:05.030 --> 00:34:05.040 align:start position:0% first the amount of work and research 00:34:05.040 --> 00:34:06.830 align:start position:0% first the amount of work and research that<00:34:05.070> goes<00:34:05.580> into<00:34:05.640> designing<00:34:06.330> an<00:34:06.510> algorithm 00:34:06.830 --> 00:34:06.840 align:start position:0% that goes into designing an algorithm 00:34:06.840 --> 00:34:09.380 align:start position:0% that goes into designing an algorithm both<00:34:07.800> AMD<00:34:08.220> and<00:34:08.340> NVIDIA<00:34:08.370> have<00:34:08.940> way<00:34:09.210> better 00:34:09.380 --> 00:34:09.390 align:start position:0% both AMD and NVIDIA have way better 00:34:09.390 --> 00:34:10.970 align:start position:0% both AMD and NVIDIA have way better things<00:34:09.659> to<00:34:09.810> do 00:34:10.970 --> 00:34:10.980 align:start position:0% things to do 00:34:10.980 --> 00:34:14.020 align:start position:0% things to do for<00:34:11.669> a<00:34:11.700> start<00:34:12.000> cryptocurrency<00:34:12.720> mining<00:34:12.990> and 00:34:14.020 --> 00:34:14.030 align:start position:0% for a start cryptocurrency mining and 00:34:14.030 --> 00:34:17.630 align:start position:0% for a start cryptocurrency mining and I'm<00:34:15.030> really<00:34:15.570> really<00:34:15.960> really<00:34:16.640> knowledgeable 00:34:17.630 --> 00:34:17.640 align:start position:0% I'm really really really knowledgeable 00:34:17.640 --> 00:34:19.340 align:start position:0% I'm really really really knowledgeable on<00:34:17.760> this<00:34:17.940> considering<00:34:18.540> I<00:34:18.659> contributed<00:34:19.230> to 00:34:19.340 --> 00:34:19.350 align:start position:0% on this considering I contributed to 00:34:19.350 --> 00:34:22.970 align:start position:0% on this considering I contributed to most<00:34:19.560> of<00:34:19.710> it<00:34:19.830> only<00:34:20.730> made<00:34:21.060> about<00:34:21.649> 384<00:34:22.649> million 00:34:22.970 --> 00:34:22.980 align:start position:0% most of it only made about 384 million 00:34:22.980 --> 00:34:26.480 align:start position:0% most of it only made about 384 million dollars<00:34:23.659> last<00:34:24.659> year<00:34:24.929> for<00:34:25.530> them<00:34:25.649> and<00:34:25.860> then<00:34:26.370> it 00:34:26.480 --> 00:34:26.490 align:start position:0% dollars last year for them and then it 00:34:26.490 --> 00:34:28.940 align:start position:0% dollars last year for them and then it made<00:34:26.700> a<00:34:26.909> tiny<00:34:27.270> amount<00:34:27.600> of<00:34:27.810> money<00:34:28.110> this<00:34:28.890> year 00:34:28.940 --> 00:34:28.950 align:start position:0% made a tiny amount of money this year 00:34:28.950 --> 00:34:30.770 align:start position:0% made a tiny amount of money this year and<00:34:29.310> you<00:34:29.399> can<00:34:29.550> publicly<00:34:30.000> track<00:34:30.300> that<00:34:30.510> because 00:34:30.770 --> 00:34:30.780 align:start position:0% and you can publicly track that because 00:34:30.780 --> 00:34:32.210 align:start position:0% and you can publicly track that because it's<00:34:30.929> all<00:34:31.020> in<00:34:31.140> their<00:34:31.290> financial<00:34:31.649> reports 00:34:32.210 --> 00:34:32.220 align:start position:0% it's all in their financial reports 00:34:32.220 --> 00:34:36.040 align:start position:0% it's all in their financial reports perks<00:34:32.850> of<00:34:33.000> a<00:34:33.060> publicly<00:34:33.480> traded<00:34:33.510> company<00:34:34.220> now 00:34:36.040 --> 00:34:36.050 align:start position:0% perks of a publicly traded company now 00:34:36.050 --> 00:34:39.890 align:start position:0% perks of a publicly traded company now their<00:34:37.050> actual<00:34:37.590> revenue<00:34:38.120> per 00:34:39.890 --> 00:34:39.900 align:start position:0% their actual revenue per 00:34:39.900 --> 00:34:42.320 align:start position:0% their actual revenue per four<00:34:40.260> GPUs<00:34:40.770> or<00:34:40.950> per<00:34:41.250> quarter<00:34:41.580> I<00:34:41.610> should<00:34:41.880> say<00:34:42.090> is 00:34:42.320 --> 00:34:42.330 align:start position:0% four GPUs or per quarter I should say is 00:34:42.330 --> 00:34:45.710 align:start position:0% four GPUs or per quarter I should say is somewhere<00:34:43.080> in<00:34:43.320> the<00:34:43.530> five<00:34:43.950> plus<00:34:44.370> billion<00:34:44.970> which 00:34:45.710 --> 00:34:45.720 align:start position:0% somewhere in the five plus billion which 00:34:45.720 --> 00:34:48.350 align:start position:0% somewhere in the five plus billion which is<00:34:45.870> a<00:34:45.900> lot<00:34:46.140> it's<00:34:46.560> a<00:34:46.710> lot<00:34:46.760> see<00:34:47.760> they<00:34:47.880> don't<00:34:48.060> care 00:34:48.350 --> 00:34:48.360 align:start position:0% is a lot it's a lot see they don't care 00:34:48.360 --> 00:34:50.570 align:start position:0% is a lot it's a lot see they don't care about<00:34:48.480> GPU<00:34:49.020> mining<00:34:49.440> they're<00:34:50.100> fighting<00:34:50.490> the 00:34:50.570 --> 00:34:50.580 align:start position:0% about GPU mining they're fighting the 00:34:50.580 --> 00:34:53.420 align:start position:0% about GPU mining they're fighting the battle<00:34:50.910> in<00:34:51.060> the<00:34:51.360> AI<00:34:51.480> space<00:34:51.990> in<00:34:52.290> the<00:34:52.650> TPU<00:34:53.010> space 00:34:53.420 --> 00:34:53.430 align:start position:0% battle in the AI space in the TPU space 00:34:53.430 --> 00:34:56.170 align:start position:0% battle in the AI space in the TPU space in<00:34:53.730> the<00:34:53.820> CPU<00:34:54.150> space<00:34:54.540> and<00:34:54.840> in<00:34:54.960> the<00:34:55.080> FPGA<00:34:55.530> space 00:34:56.170 --> 00:34:56.180 align:start position:0% in the CPU space and in the FPGA space 00:34:56.180 --> 00:34:59.840 align:start position:0% in the CPU space and in the FPGA space the<00:34:57.180> other<00:34:57.330> thing<00:34:57.630> is<00:34:57.810> that<00:34:57.840> if<00:34:58.710> AMD<00:34:59.520> had 00:34:59.840 --> 00:34:59.850 align:start position:0% the other thing is that if AMD had 00:34:59.850 --> 00:35:02.060 align:start position:0% the other thing is that if AMD had designed<00:35:00.240> this<00:35:00.450> they<00:35:01.290> would<00:35:01.530> probably<00:35:01.860> have 00:35:02.060 --> 00:35:02.070 align:start position:0% designed this they would probably have 00:35:02.070 --> 00:35:05.360 align:start position:0% designed this they would probably have done<00:35:02.520> a<00:35:02.550> better<00:35:02.880> job<00:35:03.060> at<00:35:03.600> fixing<00:35:04.140> their<00:35:04.830> shitty 00:35:05.360 --> 00:35:05.370 align:start position:0% done a better job at fixing their shitty 00:35:05.370 --> 00:35:08.300 align:start position:0% done a better job at fixing their shitty OpenCL<00:35:06.140> implementation<00:35:07.140> and<00:35:07.350> they<00:35:08.040> wouldn't 00:35:08.300 --> 00:35:08.310 align:start position:0% OpenCL implementation and they wouldn't 00:35:08.310 --> 00:35:10.460 align:start position:0% OpenCL implementation and they wouldn't have<00:35:08.490> given<00:35:08.970> me<00:35:09.240> so<00:35:09.420> much<00:35:09.540> hell<00:35:09.810> in<00:35:10.080> trying<00:35:10.350> to 00:35:10.460 --> 00:35:10.470 align:start position:0% have given me so much hell in trying to 00:35:10.470 --> 00:35:13.670 align:start position:0% have given me so much hell in trying to get<00:35:10.590> fixes<00:35:11.040> and<00:35:11.400> the<00:35:12.090> other<00:35:12.240> thing<00:35:12.510> is<00:35:12.690> again 00:35:13.670 --> 00:35:13.680 align:start position:0% get fixes and the other thing is again 00:35:13.680 --> 00:35:16.190 align:start position:0% get fixes and the other thing is again the<00:35:14.610> whole<00:35:14.820> point<00:35:15.120> of<00:35:15.300> prog<00:35:15.540> Pao<00:35:15.780> is<00:35:15.810> to 00:35:16.190 --> 00:35:16.200 align:start position:0% the whole point of prog Pao is to 00:35:16.200 --> 00:35:18.710 align:start position:0% the whole point of prog Pao is to protect<00:35:16.560> at<00:35:16.770> hash<00:35:17.100> wallet<00:35:18.000> transitions<00:35:18.570> to 00:35:18.710 --> 00:35:18.720 align:start position:0% protect at hash wallet transitions to 00:35:18.720 --> 00:35:21.650 align:start position:0% protect at hash wallet transitions to proof<00:35:18.900> of<00:35:19.020> stake<00:35:19.580> we<00:35:20.580> do<00:35:20.820> not<00:35:21.030> let<00:35:21.390> me<00:35:21.510> make 00:35:21.650 --> 00:35:21.660 align:start position:0% proof of stake we do not let me make 00:35:21.660 --> 00:35:22.400 align:start position:0% proof of stake we do not let me make this<00:35:21.810> very<00:35:22.050> clear 00:35:22.400 --> 00:35:22.410 align:start position:0% this very clear 00:35:22.410 --> 00:35:24.680 align:start position:0% this very clear if<00:35:22.560> def<00:35:22.890> else<00:35:23.130> does<00:35:23.550> not<00:35:23.820> want<00:35:24.060> proof<00:35:24.300> of<00:35:24.480> work 00:35:24.680 --> 00:35:24.690 align:start position:0% if def else does not want proof of work 00:35:24.690 --> 00:35:27.640 align:start position:0% if def else does not want proof of work to<00:35:25.350> exist<00:35:25.800> when<00:35:25.980> proof<00:35:26.280> of<00:35:26.400> stake<00:35:26.640> is<00:35:26.790> ready 00:35:27.640 --> 00:35:27.650 align:start position:0% to exist when proof of stake is ready 00:35:27.650 --> 00:35:31.430 align:start position:0% to exist when proof of stake is ready the<00:35:28.650> main<00:35:28.920> reason<00:35:29.610> for<00:35:29.640> this<00:35:30.000> is<00:35:30.300> to<00:35:31.020> make<00:35:31.200> sure 00:35:31.430 --> 00:35:31.440 align:start position:0% the main reason for this is to make sure 00:35:31.440 --> 00:35:33.590 align:start position:0% the main reason for this is to make sure that<00:35:32.040> proof<00:35:32.400> of<00:35:32.550> stake<00:35:32.790> has<00:35:33.180> a<00:35:33.210> fighting 00:35:33.590 --> 00:35:33.600 align:start position:0% that proof of stake has a fighting 00:35:33.600 --> 00:35:36.620 align:start position:0% that proof of stake has a fighting chance<00:35:33.720> and<00:35:34.290> etherium<00:35:35.480> doesn't<00:35:36.480> get 00:35:36.620 --> 00:35:36.630 align:start position:0% chance and etherium doesn't get 00:35:36.630 --> 00:35:39.110 align:start position:0% chance and etherium doesn't get destroyed<00:35:37.080> by<00:35:37.710> centralized<00:35:38.430> entities<00:35:38.910> while 00:35:39.110 --> 00:35:39.120 align:start position:0% destroyed by centralized entities while 00:35:39.120 --> 00:35:42.250 align:start position:0% destroyed by centralized entities while it's<00:35:39.330> going<00:35:39.600> into<00:35:39.720> its<00:35:39.930> transition<00:35:40.440> phase<00:35:40.620> I 00:35:42.250 --> 00:35:42.260 align:start position:0% it's going into its transition phase I 00:35:42.260 --> 00:35:44.750 align:start position:0% it's going into its transition phase I just<00:35:43.260> think<00:35:43.470> that<00:35:43.620> there<00:35:43.830> is<00:35:43.950> a<00:35:43.980> conflict<00:35:44.430> of 00:35:44.750 --> 00:35:44.760 align:start position:0% just think that there is a conflict of 00:35:44.760 --> 00:35:47.480 align:start position:0% just think that there is a conflict of interest<00:35:45.390> for<00:35:45.570> you<00:35:45.900> to<00:35:46.140> pour<00:35:46.350> that<00:35:46.530> prop<00:35:47.220> pal 00:35:47.480 --> 00:35:47.490 align:start position:0% interest for you to pour that prop pal 00:35:47.490 --> 00:35:52.850 align:start position:0% interest for you to pour that prop pal right<00:35:48.360> now<00:35:48.630> why<00:35:48.900> I<00:35:50.540> can<00:35:51.540> make<00:35:51.660> an<00:35:51.810> ASIC<00:35:52.230> I'm 00:35:52.850 --> 00:35:52.860 align:start position:0% right now why I can make an ASIC I'm 00:35:52.860 --> 00:35:56.120 align:start position:0% right now why I can make an ASIC I'm like<00:35:53.670> let's<00:35:53.910> be<00:35:54.090> very<00:35:54.510> very<00:35:54.780> public<00:35:55.020> prog<00:35:55.830> pal 00:35:56.120 --> 00:35:56.130 align:start position:0% like let's be very very public prog pal 00:35:56.130 --> 00:35:58.820 align:start position:0% like let's be very very public prog pal destroys<00:35:56.790> every<00:35:57.480> single<00:35:57.870> optimization<00:35:58.680> then 00:35:58.820 --> 00:35:58.830 align:start position:0% destroys every single optimization then 00:35:58.830 --> 00:36:01.490 align:start position:0% destroys every single optimization then I<00:35:58.890> get<00:35:59.070> pocket<00:35:59.700> money<00:35:59.730> for<00:36:00.120> this<00:36:00.930> destroys<00:36:01.350> a 00:36:01.490 --> 00:36:01.500 align:start position:0% I get pocket money for this destroys a 00:36:01.500 --> 00:36:04.190 align:start position:0% I get pocket money for this destroys a lodgment<00:36:01.850> this<00:36:02.850> destroys<00:36:03.300> all<00:36:03.450> my<00:36:03.600> private 00:36:04.190 --> 00:36:04.200 align:start position:0% lodgment this destroys all my private 00:36:04.200 --> 00:36:06.650 align:start position:0% lodgment this destroys all my private optimizations<00:36:05.190> this<00:36:05.670> destroys<00:36:06.120> the<00:36:06.270> acorn 00:36:06.650 --> 00:36:06.660 align:start position:0% optimizations this destroys the acorn 00:36:06.660 --> 00:36:09.500 align:start position:0% optimizations this destroys the acorn line<00:36:06.870> of<00:36:07.020> FPGAs<00:36:07.620> which<00:36:08.370> I<00:36:08.670> think<00:36:09.000> some<00:36:09.300> of<00:36:09.330> you 00:36:09.500 --> 00:36:09.510 align:start position:0% line of FPGAs which I think some of you 00:36:09.510 --> 00:36:11.150 align:start position:0% line of FPGAs which I think some of you are<00:36:09.540> familiar<00:36:09.810> with<00:36:10.170> the<00:36:10.380> minority<00:36:10.830> group<00:36:11.010> is 00:36:11.150 --> 00:36:11.160 align:start position:0% are familiar with the minority group is 00:36:11.160 --> 00:36:15.320 align:start position:0% are familiar with the minority group is selling<00:36:11.490> and<00:36:11.670> we<00:36:12.000> do<00:36:12.150> very<00:36:12.570> much<00:36:12.810> support<00:36:14.330> this 00:36:15.320 --> 00:36:15.330 align:start position:0% selling and we do very much support this 00:36:15.330 --> 00:36:17.810 align:start position:0% selling and we do very much support this destroys<00:36:15.810> a<00:36:16.200> lot<00:36:16.620> of<00:36:16.740> things<00:36:16.950> that<00:36:17.520> we<00:36:17.610> very 00:36:17.810 --> 00:36:17.820 align:start position:0% destroys a lot of things that we very 00:36:17.820 --> 00:36:26.920 align:start position:0% destroys a lot of things that we very much<00:36:18.030> like 00:36:26.920 --> 00:36:26.930 align:start position:0% 00:36:26.930 --> 00:36:30.530 align:start position:0% hey<00:36:27.930> I<00:36:27.960> just<00:36:28.050> curious<00:36:28.650> about<00:36:28.680> the<00:36:29.540> statistics 00:36:30.530 --> 00:36:30.540 align:start position:0% hey I just curious about the statistics 00:36:30.540 --> 00:36:32.630 align:start position:0% hey I just curious about the statistics and<00:36:30.780> security<00:36:31.350> of<00:36:31.560> the<00:36:31.710> random<00:36:32.010> masts<00:36:32.220> how<00:36:32.580> do 00:36:32.630 --> 00:36:32.640 align:start position:0% and security of the random masts how do 00:36:32.640 --> 00:36:34.640 align:start position:0% and security of the random masts how do you<00:36:32.820> prove<00:36:33.090> that<00:36:33.150> it<00:36:33.570> can<00:36:34.080> be<00:36:34.170> performed<00:36:34.350> much 00:36:34.640 --> 00:36:34.650 align:start position:0% you prove that it can be performed much 00:36:34.650 --> 00:36:38.000 align:start position:0% you prove that it can be performed much better<00:36:34.680> than<00:36:35.210> FME<00:36:36.410> because<00:36:37.410> I<00:36:37.590> checked<00:36:37.890> this 00:36:38.000 --> 00:36:38.010 align:start position:0% better than FME because I checked this 00:36:38.010 --> 00:36:40.730 align:start position:0% better than FME because I checked this back<00:36:38.340> it's<00:36:38.670> just<00:36:38.790> some<00:36:39.210> random<00:36:39.740> mathematics 00:36:40.730 --> 00:36:40.740 align:start position:0% back it's just some random mathematics 00:36:40.740 --> 00:36:42.740 align:start position:0% back it's just some random mathematics calculation<00:36:41.340> based<00:36:41.610> on<00:36:41.730> the<00:36:41.790> ha<00:36:42.030> pill<00:36:42.630> is 00:36:42.740 --> 00:36:42.750 align:start position:0% calculation based on the ha pill is 00:36:42.750 --> 00:36:44.570 align:start position:0% calculation based on the ha pill is calculated<00:36:43.320> value<00:36:43.590> if<00:36:43.950> my<00:36:44.130> understand<00:36:44.490> is 00:36:44.570 --> 00:36:44.580 align:start position:0% calculated value if my understand is 00:36:44.580 --> 00:36:47.420 align:start position:0% calculated value if my understand is correct<00:36:44.970> a<00:36:45.150> previous<00:36:45.870> previous<00:36:46.530> block<00:36:46.770> block 00:36:47.420 --> 00:36:47.430 align:start position:0% correct a previous previous block block 00:36:47.430 --> 00:36:50.210 align:start position:0% correct a previous previous block block header<00:36:47.670> not<00:36:48.660> the<00:36:48.750> previous<00:36:48.930> value<00:36:49.440> that's<00:36:49.890> a 00:36:50.210 --> 00:36:50.220 align:start position:0% header not the previous value that's a 00:36:50.220 --> 00:36:54.020 align:start position:0% header not the previous value that's a that's<00:36:50.790> a<00:36:50.970> mistake<00:36:51.350> so<00:36:52.910> specifically<00:36:53.910> the 00:36:54.020 --> 00:36:54.030 align:start position:0% that's a mistake so specifically the 00:36:54.030 --> 00:36:57.950 align:start position:0% that's a mistake so specifically the random<00:36:54.510> math<00:36:54.750> function<00:36:55.760> takes<00:36:56.760> a<00:36:57.150> few<00:36:57.510> few 00:36:57.950 --> 00:36:57.960 align:start position:0% random math function takes a few few 00:36:57.960 --> 00:37:00.140 align:start position:0% random math function takes a few few parts<00:36:58.230> it<00:36:58.350> takes<00:36:58.380> kiss<00:36:58.890> nines<00:36:59.400> random<00:36:59.970> math 00:37:00.140 --> 00:37:00.150 align:start position:0% parts it takes kiss nines random math 00:37:00.150 --> 00:37:02.330 align:start position:0% parts it takes kiss nines random math generator<00:37:00.630> it<00:37:00.900> takes<00:37:01.560> the<00:37:01.710> random<00:37:02.070> block 00:37:02.330 --> 00:37:02.340 align:start position:0% generator it takes the random block 00:37:02.340 --> 00:37:05.420 align:start position:0% generator it takes the random block header<00:37:02.760> that<00:37:03.360> changes<00:37:03.900> a<00:37:04.110> random<00:37:05.070> block<00:37:05.220> hash 00:37:05.420 --> 00:37:05.430 align:start position:0% header that changes a random block hash 00:37:05.430 --> 00:37:08.780 align:start position:0% header that changes a random block hash sorry<00:37:05.760> and<00:37:06.000> it<00:37:06.630> also<00:37:06.780> changes<00:37:07.430> the<00:37:08.430> input 00:37:08.780 --> 00:37:08.790 align:start position:0% sorry and it also changes the input 00:37:08.790 --> 00:37:12.890 align:start position:0% sorry and it also changes the input changes<00:37:09.180> every<00:37:09.540> 50<00:37:10.020> blocks<00:37:11.570> now<00:37:12.570> it's 00:37:12.890 --> 00:37:12.900 align:start position:0% changes every 50 blocks now it's 00:37:12.900 --> 00:37:15.650 align:start position:0% changes every 50 blocks now it's tweakable<00:37:13.790> where<00:37:14.790> you<00:37:14.880> can<00:37:15.060> adjust<00:37:15.210> it<00:37:15.630> to 00:37:15.650 --> 00:37:15.660 align:start position:0% tweakable where you can adjust it to 00:37:15.660 --> 00:37:18.470 align:start position:0% tweakable where you can adjust it to change<00:37:16.110> every<00:37:16.440> 25<00:37:16.860> blocks<00:37:17.070> instead<00:37:17.550> or<00:37:18.180> every 00:37:18.470 --> 00:37:18.480 align:start position:0% change every 25 blocks instead or every 00:37:18.480 --> 00:37:19.880 align:start position:0% change every 25 blocks instead or every 5<00:37:18.720> blocks<00:37:19.020> if<00:37:19.200> you<00:37:19.230> really<00:37:19.590> want<00:37:19.740> to<00:37:19.800> be 00:37:19.880 --> 00:37:19.890 align:start position:0% 5 blocks if you really want to be 00:37:19.890 --> 00:37:20.350 align:start position:0% 5 blocks if you really want to be painful 00:37:20.350 --> 00:37:20.360 align:start position:0% painful 00:37:20.360 --> 00:37:24.950 align:start position:0% painful it's<00:37:21.360> FN<00:37:21.920> v1a<00:37:22.920> that<00:37:23.190> we<00:37:23.310> use<00:37:23.550> not<00:37:23.790> FNV<00:37:24.360> itself 00:37:24.950 --> 00:37:24.960 align:start position:0% it's FN v1a that we use not FNV itself 00:37:24.960 --> 00:37:28.550 align:start position:0% it's FN v1a that we use not FNV itself like<00:37:25.680> I<00:37:25.830> said<00:37:26.070> it<00:37:26.640> takes<00:37:26.820> C<00:37:27.660> dag<00:37:27.870> which<00:37:28.350> is<00:37:28.530> a 00:37:28.550 --> 00:37:28.560 align:start position:0% like I said it takes C dag which is a 00:37:28.560 --> 00:37:30.710 align:start position:0% like I said it takes C dag which is a combination<00:37:29.220> of<00:37:29.370> your<00:37:29.850> case<00:37:30.150> your<00:37:30.510> block 00:37:30.710 --> 00:37:30.720 align:start position:0% combination of your case your block 00:37:30.720 --> 00:37:34.280 align:start position:0% combination of your case your block header<00:37:30.960> and<00:37:31.610> some<00:37:32.610> random<00:37:32.970> math<00:37:33.210> mixes<00:37:34.170> up 00:37:34.280 --> 00:37:34.290 align:start position:0% header and some random math mixes up 00:37:34.290 --> 00:37:36.320 align:start position:0% header and some random math mixes up with<00:37:34.470> more<00:37:34.680> random<00:37:35.040> math<00:37:35.250> keeps<00:37:35.880> doing<00:37:36.180> that 00:37:36.320 --> 00:37:36.330 align:start position:0% with more random math keeps doing that 00:37:36.330 --> 00:37:40.790 align:start position:0% with more random math keeps doing that for<00:37:37.140> about<00:37:37.230> I<00:37:37.890> think<00:37:38.430> we're<00:37:38.610> doing<00:37:38.790> it<00:37:39.000> 64<00:37:39.800> 64 00:37:40.790 --> 00:37:40.800 align:start position:0% for about I think we're doing it 64 64 00:37:40.800 --> 00:37:45.920 align:start position:0% for about I think we're doing it 64 64 cycles<00:37:41.270> then<00:37:42.270> spits<00:37:42.570> it<00:37:42.690> out<00:37:44.090> the<00:37:45.090> code<00:37:45.330> the 00:37:45.920 --> 00:37:45.930 align:start position:0% cycles then spits it out the code the 00:37:45.930 --> 00:37:54.910 align:start position:0% cycles then spits it out the code the code<00:37:46.080> is<00:37:46.290> the<00:37:46.410> proof 00:37:54.910 --> 00:37:54.920 align:start position:0% 00:37:54.920 --> 00:37:59.380 align:start position:0% mm-hmm<00:37:56.769> yeah<00:37:57.769> you<00:37:57.829> can<00:37:58.220> go<00:37:58.519> and<00:37:58.730> grab<00:37:58.970> the<00:37:59.150> code 00:37:59.380 --> 00:37:59.390 align:start position:0% mm-hmm yeah you can go and grab the code 00:37:59.390 --> 00:38:01.390 align:start position:0% mm-hmm yeah you can go and grab the code today<00:37:59.660> and<00:38:00.109> you<00:38:00.260> can<00:38:00.410> profile<00:38:00.829> it<00:38:00.950> yourself 00:38:01.390 --> 00:38:01.400 align:start position:0% today and you can profile it yourself 00:38:01.400 --> 00:38:03.549 align:start position:0% today and you can profile it yourself usually<00:38:02.390> through<00:38:02.599> synopsis<00:38:03.170> or<00:38:03.349> another 00:38:03.549 --> 00:38:03.559 align:start position:0% usually through synopsis or another 00:38:03.559 --> 00:38:11.289 align:start position:0% usually through synopsis or another simulator<00:38:09.190> again<00:38:10.190> all<00:38:10.519> in<00:38:10.819> the<00:38:10.910> white<00:38:11.029> paper 00:38:11.289 --> 00:38:11.299 align:start position:0% simulator again all in the white paper 00:38:11.299 --> 00:38:14.200 align:start position:0% simulator again all in the white paper yes<00:38:11.809> everything's<00:38:12.740> completely<00:38:13.279> public<00:38:14.059> with 00:38:14.200 --> 00:38:14.210 align:start position:0% yes everything's completely public with 00:38:14.210 --> 00:38:16.029 align:start position:0% yes everything's completely public with that<00:38:14.240> we<00:38:14.779> actually<00:38:14.900> did<00:38:15.230> a<00:38:15.260> huge<00:38:15.470> ride<00:38:15.740> up<00:38:15.890> on 00:38:16.029 --> 00:38:16.039 align:start position:0% that we actually did a huge ride up on 00:38:16.039 --> 00:38:17.650 align:start position:0% that we actually did a huge ride up on medium<00:38:16.250> that<00:38:16.549> we<00:38:16.670> published<00:38:17.059> two<00:38:17.150> days<00:38:17.420> ago 00:38:17.650 --> 00:38:17.660 align:start position:0% medium that we published two days ago 00:38:17.660 --> 00:38:21.789 align:start position:0% medium that we published two days ago about<00:38:17.779> this<00:38:18.490> about<00:38:19.490> the<00:38:19.700> security<00:38:20.299> of<00:38:20.450> the<00:38:21.380> FN 00:38:21.789 --> 00:38:21.799 align:start position:0% about this about the security of the FN 00:38:21.799 --> 00:38:29.710 align:start position:0% about this about the security of the FN v1<00:38:22.279> a<00:38:22.309> function<00:38:25.240> yep<00:38:28.269> just<00:38:29.269> to<00:38:29.329> the<00:38:29.420> point<00:38:29.630> you 00:38:29.710 --> 00:38:29.720 align:start position:0% v1 a function yep just to the point you 00:38:29.720 --> 00:38:33.579 align:start position:0% v1 a function yep just to the point you made<00:38:29.869> earlier<00:38:31.569> so<00:38:32.569> this<00:38:32.720> is<00:38:32.869> purely<00:38:33.289> an 00:38:33.579 --> 00:38:33.589 align:start position:0% made earlier so this is purely an 00:38:33.589 --> 00:38:37.720 align:start position:0% made earlier so this is purely an altruistic<00:38:34.099> effort<00:38:34.940> by<00:38:35.299> FFLs<00:38:35.839> to<00:38:36.829> transition 00:38:37.720 --> 00:38:37.730 align:start position:0% altruistic effort by FFLs to transition 00:38:37.730 --> 00:38:41.829 align:start position:0% altruistic effort by FFLs to transition safely<00:38:38.150> to<00:38:38.359> proof<00:38:38.750> estate<00:38:39.079> god<00:38:39.619> yes<00:38:39.950> i'm<00:38:40.839> of 00:38:41.829 --> 00:38:41.839 align:start position:0% safely to proof estate god yes i'm of 00:38:41.839 --> 00:38:43.870 align:start position:0% safely to proof estate god yes i'm of course<00:38:41.900> course<00:38:42.589> scientific<00:38:43.190> love's<00:38:43.460> a<00:38:43.609> six 00:38:43.870 --> 00:38:43.880 align:start position:0% course course scientific love's a six 00:38:43.880 --> 00:38:45.880 align:start position:0% course course scientific love's a six please<00:38:44.480> we<00:38:44.839> want<00:38:45.049> a<00:38:45.140> six<00:38:45.410> to<00:38:45.529> be<00:38:45.650> around 00:38:45.880 --> 00:38:45.890 align:start position:0% please we want a six to be around 00:38:45.890 --> 00:38:47.589 align:start position:0% please we want a six to be around because<00:38:46.160> they<00:38:46.279> consume<00:38:46.579> more<00:38:46.789> power<00:38:47.000> it's 00:38:47.589 --> 00:38:47.599 align:start position:0% because they consume more power it's 00:38:47.599 --> 00:38:53.319 align:start position:0% because they consume more power it's better<00:38:47.839> for<00:38:48.019> me<00:38:48.170> and<00:38:51.369> deaf<00:38:52.369> and<00:38:52.579> else<00:38:52.700> couldn't 00:38:53.319 --> 00:38:53.329 align:start position:0% better for me and deaf and else couldn't 00:38:53.329 --> 00:38:56.019 align:start position:0% better for me and deaf and else couldn't really<00:38:54.140> care<00:38:54.589> less<00:38:54.859> right<00:38:55.099> now<00:38:55.160> they're<00:38:55.640> GPU 00:38:56.019 --> 00:38:56.029 align:start position:0% really care less right now they're GPU 00:38:56.029 --> 00:38:58.390 align:start position:0% really care less right now they're GPU enthusiasts<00:38:56.809> we<00:38:57.380> think<00:38:57.619> it's<00:38:57.740> really<00:38:58.010> cool 00:38:58.390 --> 00:38:58.400 align:start position:0% enthusiasts we think it's really cool 00:38:58.400 --> 00:38:59.980 align:start position:0% enthusiasts we think it's really cool we<00:38:58.549> like<00:38:58.730> algorithms<00:38:59.269> a<00:38:59.390> lot<00:38:59.599> we<00:38:59.779> like 00:38:59.980 --> 00:38:59.990 align:start position:0% we like algorithms a lot we like 00:38:59.990 --> 00:39:05.849 align:start position:0% we like algorithms a lot we like proof-of-work<00:39:00.380> a<00:39:00.650> lot<00:39:00.799> it's<00:39:01.130> kind<00:39:01.369> of<00:39:01.430> fun<00:39:04.690> I 00:39:05.849 --> 00:39:05.859 align:start position:0% proof-of-work a lot it's kind of fun I 00:39:05.859 --> 00:39:08.859 align:start position:0% proof-of-work a lot it's kind of fun I noticed<00:39:06.859> that<00:39:07.069> someone<00:39:07.519> on<00:39:07.880> Bitcoin<00:39:08.299> talk 00:39:08.859 --> 00:39:08.869 align:start position:0% noticed that someone on Bitcoin talk 00:39:08.869 --> 00:39:11.730 align:start position:0% noticed that someone on Bitcoin talk claimed<00:39:09.589> that<00:39:09.980> they<00:39:10.430> found<00:39:10.759> a<00:39:10.970> way<00:39:11.000> to 00:39:11.730 --> 00:39:11.740 align:start position:0% claimed that they found a way to 00:39:11.740 --> 00:39:16.960 align:start position:0% claimed that they found a way to represent<00:39:12.740> the<00:39:12.920> food<00:39:13.690> Wawa<00:39:15.460> yes<00:39:16.460> the<00:39:16.730> food 00:39:16.960 --> 00:39:16.970 align:start position:0% represent the food Wawa yes the food 00:39:16.970 --> 00:39:19.749 align:start position:0% represent the food Wawa yes the food Aggie<00:39:17.299> yeah<00:39:17.960> several<00:39:18.230> megabytes<00:39:18.680> so<00:39:19.250> what 00:39:19.749 --> 00:39:19.759 align:start position:0% Aggie yeah several megabytes so what 00:39:19.759 --> 00:39:25.630 align:start position:0% Aggie yeah several megabytes so what that<00:39:19.970> well<00:39:20.269> if<00:39:20.990> he<00:39:21.500> is<00:39:21.559> not<00:39:22.009> lying<00:39:22.390> what<00:39:24.640> the 00:39:25.630 --> 00:39:25.640 align:start position:0% that well if he is not lying what the 00:39:25.640 --> 00:39:28.630 align:start position:0% that well if he is not lying what the protocol<00:39:26.059> I<00:39:26.329> will<00:39:27.289> again<00:39:27.859> we<00:39:28.160> pointed<00:39:28.549> that 00:39:28.630 --> 00:39:28.640 align:start position:0% protocol I will again we pointed that 00:39:28.640 --> 00:39:31.839 align:start position:0% protocol I will again we pointed that out<00:39:28.700> it's<00:39:29.000> the<00:39:29.180> FN<00:39:29.480> v1a<00:39:29.930> attack<00:39:30.410> it<00:39:31.130> does<00:39:31.369> or<00:39:31.670> F 00:39:31.839 --> 00:39:31.849 align:start position:0% out it's the FN v1a attack it does or F 00:39:31.849 --> 00:39:34.710 align:start position:0% out it's the FN v1a attack it does or F and<00:39:32.059> V<00:39:32.119> attack<00:39:32.539> it<00:39:32.750> doesn't<00:39:33.380> affect<00:39:33.470> prog<00:39:33.920> pal 00:39:34.710 --> 00:39:34.720 align:start position:0% and V attack it doesn't affect prog pal 00:39:34.720 --> 00:39:37.900 align:start position:0% and V attack it doesn't affect prog pal but<00:39:35.720> yes<00:39:35.930> so<00:39:36.230> Wawa<00:39:36.589> is<00:39:36.829> on<00:39:37.130> to<00:39:37.309> something<00:39:37.700> and 00:39:37.900 --> 00:39:37.910 align:start position:0% but yes so Wawa is on to something and 00:39:37.910 --> 00:39:45.789 align:start position:0% but yes so Wawa is on to something and he<00:39:38.599> will<00:39:38.720> be<00:39:38.869> a<00:39:38.900> very<00:39:39.170> rich<00:39:39.289> man<00:39:44.170> is<00:39:45.170> there<00:39:45.559> any 00:39:45.789 --> 00:39:45.799 align:start position:0% he will be a very rich man is there any 00:39:45.799 --> 00:39:50.319 align:start position:0% he will be a very rich man is there any like<00:39:46.309> economic<00:39:46.880> analysis<00:39:47.450> on<00:39:48.190> how<00:39:49.190> that<00:39:49.329> well 00:39:50.319 --> 00:39:50.329 align:start position:0% like economic analysis on how that well 00:39:50.329 --> 00:39:53.319 align:start position:0% like economic analysis on how that well first<00:39:50.539> of<00:39:50.660> all<00:39:50.720> how<00:39:51.019> the<00:39:52.059> hardware<00:39:53.059> that<00:39:53.089> you 00:39:53.319 --> 00:39:53.329 align:start position:0% first of all how the hardware that you 00:39:53.329 --> 00:39:57.130 align:start position:0% first of all how the hardware that you can<00:39:53.480> currently<00:39:53.809> buy<00:39:54.789> improves<00:39:55.789> current<00:39:56.690> GPUs 00:39:57.130 --> 00:39:57.140 align:start position:0% can currently buy improves current GPUs 00:39:57.140 --> 00:40:00.519 align:start position:0% can currently buy improves current GPUs performance<00:39:57.740> versus<00:39:58.660> basic<00:39:59.660> miners<00:40:00.019> and<00:40:00.289> how 00:40:00.519 --> 00:40:00.529 align:start position:0% performance versus basic miners and how 00:40:00.529 --> 00:40:04.059 align:start position:0% performance versus basic miners and how would<00:40:00.589> that<00:40:01.420> like<00:40:02.420> how<00:40:02.660> would<00:40:02.809> that<00:40:02.930> change<00:40:03.200> in 00:40:04.059 --> 00:40:04.069 align:start position:0% would that like how would that change in 00:40:04.069 --> 00:40:07.240 align:start position:0% would that like how would that change in like<00:40:04.579> in<00:40:04.759> the<00:40:04.910> world<00:40:05.180> of<00:40:05.420> proc<00:40:05.690> Val<00:40:06.130> so<00:40:07.130> the 00:40:07.240 --> 00:40:07.250 align:start position:0% like in the world of proc Val so the 00:40:07.250 --> 00:40:08.890 align:start position:0% like in the world of proc Val so the biggest<00:40:07.430> thing<00:40:07.670> is<00:40:07.940> when<00:40:08.119> we<00:40:08.180> do<00:40:08.420> our 00:40:08.890 --> 00:40:08.900 align:start position:0% biggest thing is when we do our 00:40:08.900 --> 00:40:11.710 align:start position:0% biggest thing is when we do our analysis<00:40:09.650> we<00:40:09.859> compare<00:40:10.309> silicone<00:40:11.089> costs<00:40:11.569> to 00:40:11.710 --> 00:40:11.720 align:start position:0% analysis we compare silicone costs to 00:40:11.720 --> 00:40:14.109 align:start position:0% analysis we compare silicone costs to silicone<00:40:12.200> cost<00:40:12.470> so<00:40:13.279> I<00:40:13.309> never<00:40:13.430> play<00:40:13.849> in<00:40:14.029> the 00:40:14.109 --> 00:40:14.119 align:start position:0% silicone cost so I never play in the 00:40:14.119 --> 00:40:17.650 align:start position:0% silicone cost so I never play in the realm<00:40:14.329> of<00:40:14.569> like<00:40:14.809> what<00:40:15.079> what<00:40:16.069> MSRP<00:40:16.759> is<00:40:16.970> for<00:40:17.390> GPU 00:40:17.650 --> 00:40:17.660 align:start position:0% realm of like what what MSRP is for GPU 00:40:17.660 --> 00:40:19.359 align:start position:0% realm of like what what MSRP is for GPU cards<00:40:18.079> because<00:40:18.440> there<00:40:18.680> is<00:40:18.769> there<00:40:19.130> are<00:40:19.220> ways 00:40:19.359 --> 00:40:19.369 align:start position:0% cards because there is there are ways 00:40:19.369 --> 00:40:22.900 align:start position:0% cards because there is there are ways around<00:40:19.400> that<00:40:20.799> the<00:40:21.799> I'm<00:40:22.369> guessing<00:40:22.789> you're 00:40:22.900 --> 00:40:22.910 align:start position:0% around that the I'm guessing you're 00:40:22.910 --> 00:40:24.609 align:start position:0% around that the I'm guessing you're talking<00:40:23.210> about<00:40:23.329> like<00:40:23.569> the<00:40:23.720> cost<00:40:23.960> of<00:40:24.140> an<00:40:24.230> ASIC 00:40:24.609 --> 00:40:24.619 align:start position:0% talking about like the cost of an ASIC 00:40:24.619 --> 00:40:27.009 align:start position:0% talking about like the cost of an ASIC versus<00:40:24.950> a<00:40:24.980> cost<00:40:25.279> of<00:40:25.400> a<00:40:25.490> GPU<00:40:25.880> in<00:40:26.509> the<00:40:26.690> world<00:40:26.869> of 00:40:27.009 --> 00:40:27.019 align:start position:0% versus a cost of a GPU in the world of 00:40:27.019 --> 00:40:30.579 align:start position:0% versus a cost of a GPU in the world of prog<00:40:27.230> pal<00:40:27.910> yeah<00:40:28.910> well<00:40:29.359> I<00:40:29.839> know<00:40:29.990> so<00:40:30.289> that 00:40:30.579 --> 00:40:30.589 align:start position:0% prog pal yeah well I know so that 00:40:30.589 --> 00:40:33.640 align:start position:0% prog pal yeah well I know so that comparing<00:40:31.220> to<00:40:31.460> the<00:40:31.670> cost<00:40:31.940> of<00:40:32.180> ASIC<00:40:33.049> versus<00:40:33.529> the 00:40:33.640 --> 00:40:33.650 align:start position:0% comparing to the cost of ASIC versus the 00:40:33.650 --> 00:40:36.549 align:start position:0% comparing to the cost of ASIC versus the cost<00:40:33.859> of<00:40:34.130> GPU<00:40:34.819> plus<00:40:35.180> acorn<00:40:35.930> plus<00:40:36.289> all<00:40:36.470> the 00:40:36.549 --> 00:40:36.559 align:start position:0% cost of GPU plus acorn plus all the 00:40:36.559 --> 00:40:37.539 align:start position:0% cost of GPU plus acorn plus all the other<00:40:36.680> stuff<00:40:36.950> oh<00:40:37.069> yeah 00:40:37.539 --> 00:40:37.549 align:start position:0% other stuff oh yeah 00:40:37.549 --> 00:40:42.819 align:start position:0% other stuff oh yeah so<00:40:37.759> let's<00:40:38.299> so<00:40:39.049> the<00:40:39.380> cost<00:40:39.680> of<00:40:40.130> a<00:40:40.309> GPU<00:40:40.640> card<00:40:41.829> will 00:40:42.819 --> 00:40:42.829 align:start position:0% so let's so the cost of a GPU card will 00:40:42.829 --> 00:40:45.160 align:start position:0% so let's so the cost of a GPU card will talk<00:40:43.009> roughly<00:40:43.400> about<00:40:43.640> that<00:40:43.819> today<00:40:44.480> it's 00:40:45.160 --> 00:40:45.170 align:start position:0% talk roughly about that today it's 00:40:45.170 --> 00:40:48.789 align:start position:0% talk roughly about that today it's around<00:40:45.730> it's<00:40:46.730> around<00:40:47.170> $100<00:40:48.170> for<00:40:48.410> the<00:40:48.499> memory 00:40:48.789 --> 00:40:48.799 align:start position:0% around it's around $100 for the memory 00:40:48.799 --> 00:40:53.410 align:start position:0% around it's around $100 for the memory on<00:40:49.249> its<00:40:49.430> own<00:40:49.579> for<00:40:49.789> 8<00:40:49.880> gigabytes<00:40:50.329> of<00:40:50.809> gddr5<00:40:52.420> g6 00:40:53.410 --> 00:40:53.420 align:start position:0% on its own for 8 gigabytes of gddr5 g6 00:40:53.420 --> 00:40:57.069 align:start position:0% on its own for 8 gigabytes of gddr5 g6 is<00:40:53.599> around<00:40:53.920> 150<00:40:54.920> the<00:40:55.579> GPU<00:40:55.970> ASIC<00:40:56.359> itself<00:40:56.749> is 00:40:57.069 --> 00:40:57.079 align:start position:0% is around 150 the GPU ASIC itself is 00:40:57.079 --> 00:41:00.460 align:start position:0% is around 150 the GPU ASIC itself is around<00:40:57.940> $40<00:40:58.940> or<00:40:59.119> $50<00:40:59.749> depending<00:41:00.259> on<00:41:00.349> it 00:41:00.460 --> 00:41:00.470 align:start position:0% around $40 or $50 depending on it 00:41:00.470 --> 00:41:03.099 align:start position:0% around $40 or $50 depending on it whether<00:41:00.619> it's<00:41:00.769> Nvidia<00:41:01.130> or<00:41:01.190> AMD<00:41:01.279> and<00:41:02.029> the<00:41:02.630> PCB 00:41:03.099 --> 00:41:03.109 align:start position:0% whether it's Nvidia or AMD and the PCB 00:41:03.109 --> 00:41:05.950 align:start position:0% whether it's Nvidia or AMD and the PCB is<00:41:03.319> 25<00:41:03.890> bucks<00:41:04.160> and<00:41:04.579> the<00:41:04.880> heatsink<00:41:05.359> admin 00:41:05.950 --> 00:41:05.960 align:start position:0% is 25 bucks and the heatsink admin 00:41:05.960 --> 00:41:09.489 align:start position:0% is 25 bucks and the heatsink admin miscellaneous<00:41:06.950> it's<00:41:07.640> around<00:41:07.880> $15<00:41:08.569> now<00:41:09.380> an 00:41:09.489 --> 00:41:09.499 align:start position:0% miscellaneous it's around $15 now an 00:41:09.499 --> 00:41:12.910 align:start position:0% miscellaneous it's around $15 now an ASIC<00:41:10.099> an<00:41:10.400> ASIC<00:41:11.359> is<00:41:11.450> really<00:41:11.749> cheap<00:41:12.019> so<00:41:12.499> we 00:41:12.910 --> 00:41:12.920 align:start position:0% ASIC an ASIC is really cheap so we 00:41:12.920 --> 00:41:14.739 align:start position:0% ASIC an ASIC is really cheap so we talked<00:41:13.369> a<00:41:13.400> little<00:41:13.609> bit<00:41:13.789> about<00:41:13.910> the<00:41:14.150> e3<00:41:14.299> because 00:41:14.739 --> 00:41:14.749 align:start position:0% talked a little bit about the e3 because 00:41:14.749 --> 00:41:16.180 align:start position:0% talked a little bit about the e3 because it's<00:41:14.900> the<00:41:14.989> one<00:41:15.140> I<00:41:15.170> have<00:41:15.349> the<00:41:15.440> most<00:41:15.589> experience 00:41:16.180 --> 00:41:16.190 align:start position:0% it's the one I have the most experience 00:41:16.190 --> 00:41:18.489 align:start position:0% it's the one I have the most experience with<00:41:16.279> I<00:41:16.609> haven't<00:41:16.880> analyzed<00:41:17.569> ed<00:41:17.869> Lindsay's<00:41:18.259> at 00:41:18.489 --> 00:41:18.499 align:start position:0% with I haven't analyzed ed Lindsay's at 00:41:18.499 --> 00:41:22.479 align:start position:0% with I haven't analyzed ed Lindsay's at hashe<00:41:18.890> 6<00:41:19.160> but<00:41:19.910> the<00:41:20.059> e3<00:41:20.539> chips<00:41:20.930> the<00:41:21.859> DRAM<00:41:22.190> chips 00:41:22.479 --> 00:41:22.489 align:start position:0% hashe 6 but the e3 chips the DRAM chips 00:41:22.489 --> 00:41:26.140 align:start position:0% hashe 6 but the e3 chips the DRAM chips were<00:41:22.940> brought<00:41:23.210> for<00:41:23.480> $3<00:41:24.109> 50<00:41:24.470> each<00:41:25.150> that's 00:41:26.140 --> 00:41:26.150 align:start position:0% were brought for $3 50 each that's 00:41:26.150 --> 00:41:28.539 align:start position:0% were brought for $3 50 each that's actually<00:41:26.539> public<00:41:27.049> on<00:41:27.230> if<00:41:27.769> you<00:41:27.950> if<00:41:28.160> you<00:41:28.279> check 00:41:28.539 --> 00:41:28.549 align:start position:0% actually public on if you if you check 00:41:28.549 --> 00:41:32.799 align:start position:0% actually public on if you if you check their<00:41:28.789> l<00:41:29.299> PETA's<00:41:29.720> Oh<00:41:30.670> Peters<00:41:31.670> filings<00:41:32.150> you<00:41:32.450> can 00:41:32.799 --> 00:41:32.809 align:start position:0% their l PETA's Oh Peters filings you can 00:41:32.809 --> 00:41:34.660 align:start position:0% their l PETA's Oh Peters filings you can see<00:41:32.989> how<00:41:33.470> much<00:41:33.650> was<00:41:33.859> brought<00:41:34.099> in<00:41:34.249> that<00:41:34.369> certain 00:41:34.660 --> 00:41:34.670 align:start position:0% see how much was brought in that certain 00:41:34.670 --> 00:41:37.509 align:start position:0% see how much was brought in that certain period<00:41:35.180> then<00:41:35.720> the<00:41:35.869> PCB<00:41:36.349> is<00:41:36.529> only<00:41:36.710> around<00:41:36.890> 20<00:41:37.369> to 00:41:37.509 --> 00:41:37.519 align:start position:0% period then the PCB is only around 20 to 00:41:37.519 --> 00:41:40.269 align:start position:0% period then the PCB is only around 20 to 25<00:41:37.549> dollars<00:41:38.329> the<00:41:39.170> heatsink<00:41:39.680> the<00:41:39.829> sheet<00:41:40.069> metal 00:41:40.269 --> 00:41:40.279 align:start position:0% 25 dollars the heatsink the sheet metal 00:41:40.279 --> 00:41:43.059 align:start position:0% 25 dollars the heatsink the sheet metal everything<00:41:40.940> else<00:41:41.119> assumed<00:41:41.720> $30<00:41:42.319> so<00:41:42.799> there<00:41:42.950> is 00:41:43.059 --> 00:41:43.069 align:start position:0% everything else assumed $30 so there is 00:41:43.069 --> 00:41:45.160 align:start position:0% everything else assumed $30 so there is quite<00:41:43.400> a<00:41:43.430> large<00:41:43.730> gap<00:41:44.180> and<00:41:44.509> then<00:41:44.809> you<00:41:44.900> have<00:41:45.049> to 00:41:45.160 --> 00:41:45.170 align:start position:0% quite a large gap and then you have to 00:41:45.170 --> 00:41:49.329 align:start position:0% quite a large gap and then you have to factor<00:41:45.410> in<00:41:45.589> the<00:41:45.859> MSRP<00:41:46.339> as<00:41:47.269> well<00:41:47.499> now<00:41:48.499> in<00:41:48.920> a 00:41:49.329 --> 00:41:49.339 align:start position:0% factor in the MSRP as well now in a 00:41:49.339 --> 00:41:51.249 align:start position:0% factor in the MSRP as well now in a probe<00:41:49.549> pal<00:41:49.789> world<00:41:50.059> they're<00:41:50.900> going<00:41:51.170> to 00:41:51.249 --> 00:41:51.259 align:start position:0% probe pal world they're going to 00:41:51.259 --> 00:41:53.079 align:start position:0% probe pal world they're going to actually<00:41:51.410> match<00:41:51.890> on<00:41:52.160> a<00:41:52.190> silicon<00:41:52.789> and<00:41:53.059> a 00:41:53.079 --> 00:41:53.089 align:start position:0% actually match on a silicon and a 00:41:53.089 --> 00:41:56.650 align:start position:0% actually match on a silicon and a heatsink<00:41:53.869> cost<00:41:54.200> and<00:41:54.619> the<00:41:55.309> only<00:41:55.489> savings<00:41:56.210> as<00:41:56.509> I 00:41:56.650 --> 00:41:56.660 align:start position:0% heatsink cost and the only savings as I 00:41:56.660 --> 00:41:58.450 align:start position:0% heatsink cost and the only savings as I as<00:41:56.989> I<00:41:57.230> said<00:41:57.440> before<00:41:57.739> you're<00:41:57.890> going<00:41:58.099> to<00:41:58.190> get<00:41:58.339> is 00:41:58.450 --> 00:41:58.460 align:start position:0% as I said before you're going to get is 00:41:58.460 --> 00:42:00.670 align:start position:0% as I said before you're going to get is by<00:41:58.640> shaving<00:41:59.119> off<00:41:59.239> those<00:41:59.450> unused<00:41:59.960> dye<00:42:00.230> cases 00:42:00.670 --> 00:42:00.680 align:start position:0% by shaving off those unused dye cases 00:42:00.680 --> 00:42:02.650 align:start position:0% by shaving off those unused dye cases that<00:42:01.039> we<00:42:01.190> can't<00:42:01.430> really<00:42:01.609> find<00:42:01.910> it<00:42:02.029> use<00:42:02.150> for<00:42:02.390> and 00:42:02.650 --> 00:42:02.660 align:start position:0% that we can't really find it use for and 00:42:02.660 --> 00:42:05.200 align:start position:0% that we can't really find it use for and that's<00:42:03.619> gonna<00:42:03.799> save<00:42:04.160> you<00:42:04.309> around<00:42:04.609> two<00:42:05.059> to 00:42:05.200 --> 00:42:05.210 align:start position:0% that's gonna save you around two to 00:42:05.210 --> 00:42:09.700 align:start position:0% that's gonna save you around two to three<00:42:05.239> dollars<00:42:05.660> per<00:42:06.109> GPU<00:42:06.769> card<00:42:07.569> so<00:42:08.710> there's 00:42:09.700 --> 00:42:09.710 align:start position:0% three dollars per GPU card so there's 00:42:09.710 --> 00:42:12.759 align:start position:0% three dollars per GPU card so there's also<00:42:09.920> the<00:42:10.339> basic<00:42:11.089> design<00:42:11.269> costs<00:42:11.989> right<00:42:12.559> and 00:42:12.759 --> 00:42:12.769 align:start position:0% also the basic design costs right and 00:42:12.769 --> 00:42:13.400 align:start position:0% also the basic design costs right and any 00:42:13.400 --> 00:42:13.410 align:start position:0% any 00:42:13.410 --> 00:42:20.420 align:start position:0% any and<00:42:15.800> ASIC<00:42:16.800> design<00:42:17.300> R&D<00:42:18.300> I<00:42:18.630> mean<00:42:19.560> I'm<00:42:20.130> not<00:42:20.370> sure 00:42:20.420 --> 00:42:20.430 align:start position:0% and ASIC design R&D I mean I'm not sure 00:42:20.430 --> 00:42:23.300 align:start position:0% and ASIC design R&D I mean I'm not sure how<00:42:20.730> public<00:42:21.240> some<00:42:21.720> people's<00:42:22.080> costs<00:42:22.530> are<00:42:22.710> but 00:42:23.300 --> 00:42:23.310 align:start position:0% how public some people's costs are but 00:42:23.310 --> 00:42:26.570 align:start position:0% how public some people's costs are but like<00:42:23.520> for<00:42:23.850> for<00:42:24.270> a<00:42:24.300> sia<00:42:24.780> ASIC<00:42:25.320> or<00:42:25.590> for<00:42:25.860> even<00:42:26.490> a 00:42:26.570 --> 00:42:26.580 align:start position:0% like for for a sia ASIC or for even a 00:42:26.580 --> 00:42:28.790 align:start position:0% like for for a sia ASIC or for even a sha-256<00:42:27.480> ASIC<00:42:27.900> you<00:42:28.020> can<00:42:28.200> get<00:42:28.320> them<00:42:28.440> as<00:42:28.560> cheap 00:42:28.790 --> 00:42:28.800 align:start position:0% sha-256 ASIC you can get them as cheap 00:42:28.800 --> 00:42:32.750 align:start position:0% sha-256 ASIC you can get them as cheap as<00:42:29.010> 10k<00:42:29.780> R&D<00:42:30.780> cost<00:42:31.020> and<00:42:31.350> a<00:42:31.800> pro<00:42:32.010> growl<00:42:32.310> a<00:42:32.340> sick 00:42:32.750 --> 00:42:32.760 align:start position:0% as 10k R&D cost and a pro growl a sick 00:42:32.760 --> 00:42:35.000 align:start position:0% as 10k R&D cost and a pro growl a sick it's<00:42:33.000> pretty<00:42:33.570> simple<00:42:34.020> to<00:42:34.080> make<00:42:34.380> since<00:42:34.740> all<00:42:34.890> the 00:42:35.000 --> 00:42:35.010 align:start position:0% it's pretty simple to make since all the 00:42:35.010 --> 00:42:37.760 align:start position:0% it's pretty simple to make since all the code<00:42:35.250> is<00:42:35.460> open<00:42:35.490> source<00:42:35.790> and<00:42:36.300> he'll<00:42:36.530> maybe<00:42:37.530> when 00:42:37.760 --> 00:42:37.770 align:start position:0% code is open source and he'll maybe when 00:42:37.770 --> 00:42:39.230 align:start position:0% code is open source and he'll maybe when we<00:42:37.860> have<00:42:38.010> some<00:42:38.190> spare<00:42:38.430> time<00:42:38.460> we<00:42:38.820> can<00:42:38.970> just<00:42:39.150> make 00:42:39.230 --> 00:42:39.240 align:start position:0% we have some spare time we can just make 00:42:39.240 --> 00:42:40.970 align:start position:0% we have some spare time we can just make open<00:42:39.600> source<00:42:39.810> schematics<00:42:40.290> of<00:42:40.440> a<00:42:40.530> probe<00:42:40.800> pal 00:42:40.970 --> 00:42:40.980 align:start position:0% open source schematics of a probe pal 00:42:40.980 --> 00:42:43.880 align:start position:0% open source schematics of a probe pal basic<00:42:41.370> why<00:42:42.000> the<00:42:42.060> hell<00:42:42.270> not<00:42:42.450> a<00:42:42.810> better<00:42:43.650> way<00:42:43.830> to 00:42:43.880 --> 00:42:43.890 align:start position:0% basic why the hell not a better way to 00:42:43.890 --> 00:42:55.550 align:start position:0% basic why the hell not a better way to prove<00:42:44.130> our<00:42:44.280> point<00:42:45.800> thank<00:42:46.800> you<00:42:54.050> yeah<00:42:55.050> I<00:42:55.080> just 00:42:55.550 --> 00:42:55.560 align:start position:0% prove our point thank you yeah I just 00:42:55.560 --> 00:42:57.500 align:start position:0% prove our point thank you yeah I just want<00:42:55.860> to<00:42:55.950> make<00:42:56.070> a<00:42:56.100> guess<00:42:56.760> a<00:42:56.820> quick<00:42:57.060> point<00:42:57.330> for 00:42:57.500 --> 00:42:57.510 align:start position:0% want to make a guess a quick point for 00:42:57.510 --> 00:42:59.960 align:start position:0% want to make a guess a quick point for people<00:42:58.170> in<00:42:58.440> the<00:42:58.560> audience<00:42:58.590> from<00:42:59.430> this<00:42:59.730> coming 00:42:59.960 --> 00:42:59.970 align:start position:0% people in the audience from this coming 00:42:59.970 --> 00:43:01.630 align:start position:0% people in the audience from this coming from<00:43:00.090> miner<00:43:00.330> miners<00:43:00.870> mostly<00:43:01.230> just<00:43:01.440> care<00:43:01.590> about 00:43:01.630 --> 00:43:01.640 align:start position:0% from miner miners mostly just care about 00:43:01.640 --> 00:43:06.740 align:start position:0% from miner miners mostly just care about the<00:43:04.460> profit<00:43:05.460> that<00:43:05.490> they're<00:43:05.700> making<00:43:05.850> correct 00:43:06.740 --> 00:43:06.750 align:start position:0% the profit that they're making correct 00:43:06.750 --> 00:43:10.130 align:start position:0% the profit that they're making correct so<00:43:07.080> it<00:43:07.800> the<00:43:08.430> revenue<00:43:08.820> is<00:43:08.940> gonna<00:43:09.150> be<00:43:09.360> you<00:43:10.020> know 00:43:10.130 --> 00:43:10.140 align:start position:0% so it the revenue is gonna be you know 00:43:10.140 --> 00:43:13.610 align:start position:0% so it the revenue is gonna be you know how<00:43:10.320> much<00:43:10.530> coin<00:43:10.800> you're<00:43:11.040> mining<00:43:12.350> times<00:43:13.350> its 00:43:13.610 --> 00:43:13.620 align:start position:0% how much coin you're mining times its 00:43:13.620 --> 00:43:16.130 align:start position:0% how much coin you're mining times its value<00:43:13.770> you<00:43:14.160> can<00:43:14.310> sell<00:43:14.520> it<00:43:14.640> in<00:43:14.760> to<00:43:14.940> you<00:43:15.240> know<00:43:15.360> all 00:43:16.130 --> 00:43:16.140 align:start position:0% value you can sell it in to you know all 00:43:16.140 --> 00:43:19.790 align:start position:0% value you can sell it in to you know all the<00:43:16.380> miners<00:43:16.680> have<00:43:17.660> real-world<00:43:18.660> Fiat<00:43:19.320> costs 00:43:19.790 --> 00:43:19.800 align:start position:0% the miners have real-world Fiat costs 00:43:19.800 --> 00:43:21.080 align:start position:0% the miners have real-world Fiat costs they<00:43:20.100> have<00:43:20.190> to<00:43:20.250> pay<00:43:20.340> the<00:43:20.430> power<00:43:20.610> companies<00:43:20.970> in 00:43:21.080 --> 00:43:21.090 align:start position:0% they have to pay the power companies in 00:43:21.090 --> 00:43:22.130 align:start position:0% they have to pay the power companies in fiat<00:43:21.420> so<00:43:21.570> they<00:43:21.630> have<00:43:21.720> to<00:43:21.780> convert<00:43:21.990> the 00:43:22.130 --> 00:43:22.140 align:start position:0% fiat so they have to convert the 00:43:22.140 --> 00:43:24.280 align:start position:0% fiat so they have to convert the cryptography<00:43:22.560> so<00:43:23.190> at<00:43:23.280> the<00:43:23.370> end<00:43:23.490> of<00:43:23.550> the<00:43:23.610> day 00:43:24.280 --> 00:43:24.290 align:start position:0% cryptography so at the end of the day 00:43:24.290 --> 00:43:26.630 align:start position:0% cryptography so at the end of the day they<00:43:25.290> just<00:43:25.500> care<00:43:25.650> about<00:43:25.680> their<00:43:25.950> profit<00:43:26.340> margin 00:43:26.630 --> 00:43:26.640 align:start position:0% they just care about their profit margin 00:43:26.640 --> 00:43:30.590 align:start position:0% they just care about their profit margin and<00:43:26.760> so<00:43:27.420> you<00:43:28.020> know<00:43:28.520> Minh<00:43:29.520> wants<00:43:29.790> to<00:43:29.910> develop<00:43:30.180> a 00:43:30.590 --> 00:43:30.600 align:start position:0% and so you know Minh wants to develop a 00:43:30.600 --> 00:43:32.990 align:start position:0% and so you know Minh wants to develop a ASIC<00:43:31.500> for<00:43:31.830> Prague<00:43:32.010> cow<00:43:32.220> and<00:43:32.490> there's<00:43:32.700> only<00:43:32.910> a 00:43:32.990 --> 00:43:33.000 align:start position:0% ASIC for Prague cow and there's only a 00:43:33.000 --> 00:43:35.510 align:start position:0% ASIC for Prague cow and there's only a 20<00:43:33.540> or<00:43:33.570> 30<00:43:33.720> percent<00:43:34.020> speed-up<00:43:34.560> just<00:43:35.280> the 00:43:35.510 --> 00:43:35.520 align:start position:0% 20 or 30 percent speed-up just the 00:43:35.520 --> 00:43:37.910 align:start position:0% 20 or 30 percent speed-up just the development<00:43:36.000> costs<00:43:36.660> for<00:43:37.080> rolling<00:43:37.500> out<00:43:37.650> the 00:43:37.910 --> 00:43:37.920 align:start position:0% development costs for rolling out the 00:43:37.920 --> 00:43:40.460 align:start position:0% development costs for rolling out the silicon<00:43:38.430> and<00:43:38.820> you<00:43:39.570> know<00:43:39.690> building<00:43:40.260> all<00:43:40.440> that 00:43:40.460 --> 00:43:40.470 align:start position:0% silicon and you know building all that 00:43:40.470 --> 00:43:41.960 align:start position:0% silicon and you know building all that which<00:43:40.890> costs<00:43:41.250> millions<00:43:41.580> and<00:43:41.670> millions<00:43:41.910> of 00:43:41.960 --> 00:43:41.970 align:start position:0% which costs millions and millions of 00:43:41.970 --> 00:43:45.980 align:start position:0% which costs millions and millions of dollars<00:43:42.030> won't<00:43:42.960> actually<00:43:44.510> make<00:43:45.510> up<00:43:45.720> for<00:43:45.930> the 00:43:45.980 --> 00:43:45.990 align:start position:0% dollars won't actually make up for the 00:43:45.990 --> 00:43:48.410 align:start position:0% dollars won't actually make up for the gap<00:43:46.140> so<00:43:46.470> it's<00:43:47.430> not<00:43:47.520> like<00:43:47.670> you<00:43:47.790> guys<00:43:47.910> have<00:43:48.150> to<00:43:48.300> be 00:43:48.410 --> 00:43:48.420 align:start position:0% gap so it's not like you guys have to be 00:43:48.420 --> 00:43:52.550 align:start position:0% gap so it's not like you guys have to be like<00:43:48.630> just<00:43:48.990> as<00:43:49.170> fast<00:43:49.410> as<00:43:49.680> ASIC<00:43:51.050> even<00:43:52.050> if<00:43:52.170> a<00:43:52.290> six 00:43:52.550 --> 00:43:52.560 align:start position:0% like just as fast as ASIC even if a six 00:43:52.560 --> 00:43:54.350 align:start position:0% like just as fast as ASIC even if a six or<00:43:52.680> ten<00:43:52.920> or<00:43:53.040> twenty<00:43:53.250> percent<00:43:53.280> faster<00:43:53.730> it's 00:43:54.350 --> 00:43:54.360 align:start position:0% or ten or twenty percent faster it's 00:43:54.360 --> 00:43:56.330 align:start position:0% or ten or twenty percent faster it's just<00:43:54.630> not<00:43:54.780> economical<00:43:55.500> for<00:43:55.680> them<00:43:55.770> to<00:43:55.920> even<00:43:56.160> go 00:43:56.330 --> 00:43:56.340 align:start position:0% just not economical for them to even go 00:43:56.340 --> 00:43:58.670 align:start position:0% just not economical for them to even go out<00:43:56.430> and<00:43:56.610> deploy<00:43:56.820> all<00:43:57.060> that<00:43:57.240> NRI<00:43:58.080> cost<00:43:58.410> to 00:43:58.670 --> 00:43:58.680 align:start position:0% out and deploy all that NRI cost to 00:43:58.680 --> 00:44:01.460 align:start position:0% out and deploy all that NRI cost to actually<00:43:58.860> build<00:43:59.280> it<00:43:59.430> so<00:44:00.240> I<00:44:00.660> don't<00:44:01.170> think<00:44:01.350> it's 00:44:01.460 --> 00:44:01.470 align:start position:0% actually build it so I don't think it's 00:44:01.470 --> 00:44:05.360 align:start position:0% actually build it so I don't think it's like<00:44:01.590> you<00:44:01.740> guys<00:44:01.830> need<00:44:02.100> to<00:44:02.130> even<00:44:04.100> make<00:44:05.100> it<00:44:05.250> so 00:44:05.360 --> 00:44:05.370 align:start position:0% like you guys need to even make it so 00:44:05.370 --> 00:44:07.070 align:start position:0% like you guys need to even make it so that<00:44:05.520> there's<00:44:05.670> no<00:44:05.940> efficiency<00:44:06.570> speed<00:44:06.810> up<00:44:06.960> but 00:44:07.070 --> 00:44:07.080 align:start position:0% that there's no efficiency speed up but 00:44:07.080 --> 00:44:08.600 align:start position:0% that there's no efficiency speed up but it<00:44:07.200> just<00:44:07.230> needs<00:44:07.470> to<00:44:07.620> be<00:44:07.770> that<00:44:07.890> there's<00:44:08.070> you 00:44:08.600 --> 00:44:08.610 align:start position:0% it just needs to be that there's you 00:44:08.610 --> 00:44:10.310 align:start position:0% it just needs to be that there's you know<00:44:08.940> even<00:44:09.480> if<00:44:09.570> there's<00:44:09.750> a<00:44:09.810> 10<00:44:09.960> and<00:44:10.110> 20<00:44:10.290> percent 00:44:10.310 --> 00:44:10.320 align:start position:0% know even if there's a 10 and 20 percent 00:44:10.320 --> 00:44:12.350 align:start position:0% know even if there's a 10 and 20 percent speed-up<00:44:10.860> it's<00:44:11.160> good<00:44:11.370> enough<00:44:11.610> so<00:44:12.060> I<00:44:12.090> mean 00:44:12.350 --> 00:44:12.360 align:start position:0% speed-up it's good enough so I mean 00:44:12.360 --> 00:44:15.800 align:start position:0% speed-up it's good enough so I mean right<00:44:12.810> now<00:44:12.870> there's<00:44:13.320> a<00:44:13.430> there's<00:44:14.430> like<00:44:14.700> a<00:44:14.810> 7 00:44:15.800 --> 00:44:15.810 align:start position:0% right now there's a there's like a 7 00:44:15.810 --> 00:44:19.010 align:start position:0% right now there's a there's like a 7 percent<00:44:15.930> to<00:44:16.410> 12<00:44:16.800> percent<00:44:16.890> speed<00:44:17.430> up<00:44:17.580> but<00:44:18.210> we 00:44:19.010 --> 00:44:19.020 align:start position:0% percent to 12 percent speed up but we 00:44:19.020 --> 00:44:20.570 align:start position:0% percent to 12 percent speed up but we also<00:44:19.200> just<00:44:19.620> like<00:44:19.800> to<00:44:19.980> make<00:44:20.130> it<00:44:20.280> so<00:44:20.400> that 00:44:20.570 --> 00:44:20.580 align:start position:0% also just like to make it so that 00:44:20.580 --> 00:44:22.910 align:start position:0% also just like to make it so that there's<00:44:20.730> no<00:44:20.970> major<00:44:21.660> speed-up<00:44:22.200> because<00:44:22.500> a<00:44:22.740> it's 00:44:22.910 --> 00:44:22.920 align:start position:0% there's no major speed-up because a it's 00:44:22.920 --> 00:44:24.680 align:start position:0% there's no major speed-up because a it's fun<00:44:23.190> sure<00:44:23.940> yeah 00:44:24.680 --> 00:44:24.690 align:start position:0% fun sure yeah 00:44:24.690 --> 00:44:29.380 align:start position:0% fun sure yeah the<00:44:25.380> fits<00:44:25.830> with<00:44:25.980> the<00:44:26.100> chaotic<00:44:26.400> theme 00:44:29.380 --> 00:44:29.390 align:start position:0% 00:44:29.390 --> 00:44:33.050 align:start position:0% just<00:44:30.390> a<00:44:30.480> follow-on<00:44:30.750> question<00:44:31.700> for<00:44:32.700> that<00:44:32.820> is 00:44:33.050 --> 00:44:33.060 align:start position:0% just a follow-on question for that is 00:44:33.060 --> 00:44:36.670 align:start position:0% just a follow-on question for that is that<00:44:33.470> presumably<00:44:34.550> prop<00:44:35.550> power<00:44:35.760> works<00:44:36.000> on 00:44:36.670 --> 00:44:36.680 align:start position:0% that presumably prop power works on 00:44:36.680 --> 00:44:39.980 align:start position:0% that presumably prop power works on standard<00:44:37.680> commodity<00:44:38.840> GPOs 00:44:39.980 --> 00:44:39.990 align:start position:0% standard commodity GPOs 00:44:39.990 --> 00:44:43.310 align:start position:0% standard commodity GPOs which<00:44:40.140> can<00:44:40.380> be<00:44:40.500> redeployed<00:44:41.540> on<00:44:42.540> other<00:44:42.900> uses 00:44:43.310 --> 00:44:43.320 align:start position:0% which can be redeployed on other uses 00:44:43.320 --> 00:44:46.070 align:start position:0% which can be redeployed on other uses like<00:44:43.350> artificial<00:44:43.860> intelligence<00:44:44.550> even<00:44:45.300> after 00:44:46.070 --> 00:44:46.080 align:start position:0% like artificial intelligence even after 00:44:46.080 --> 00:44:48.859 align:start position:0% like artificial intelligence even after a<00:44:46.350> theorem<00:44:46.920> goes<00:44:47.100> to<00:44:47.340> proof<00:44:47.610> of<00:44:47.760> stake<00:44:48.030> that's 00:44:48.859 --> 00:44:48.869 align:start position:0% a theorem goes to proof of stake that's 00:44:48.869 --> 00:44:50.510 align:start position:0% a theorem goes to proof of stake that's another<00:44:49.290> very<00:44:49.650> big<00:44:49.860> factor<00:44:50.130> if<00:44:50.400> you're 00:44:50.510 --> 00:44:50.520 align:start position:0% another very big factor if you're 00:44:50.520 --> 00:44:52.490 align:start position:0% another very big factor if you're considering<00:44:50.970> deploying<00:44:51.450> tens<00:44:52.020> or<00:44:52.200> hundreds 00:44:52.490 --> 00:44:52.500 align:start position:0% considering deploying tens or hundreds 00:44:52.500 --> 00:44:55.160 align:start position:0% considering deploying tens or hundreds of<00:44:52.590> millions<00:44:52.860> of<00:44:52.950> dollars<00:44:53.100> on<00:44:53.369> an<00:44:53.490> ASIC<00:44:54.170> which 00:44:55.160 --> 00:44:55.170 align:start position:0% of millions of dollars on an ASIC which 00:44:55.170 --> 00:44:57.830 align:start position:0% of millions of dollars on an ASIC which has<00:44:55.320> only<00:44:55.530> got<00:44:55.800> one<00:44:55.830> purpose<00:44:56.460> where<00:44:57.240> as<00:44:57.390> a<00:44:57.420> GPU 00:44:57.830 --> 00:44:57.840 align:start position:0% has only got one purpose where as a GPU 00:44:57.840 --> 00:45:02.570 align:start position:0% has only got one purpose where as a GPU farm<00:44:58.740> can<00:44:59.520> be<00:44:59.640> reused<00:45:00.080> is<00:45:01.080> that<00:45:01.290> the<00:45:01.410> case<00:45:01.590> that 00:45:02.570 --> 00:45:02.580 align:start position:0% farm can be reused is that the case that 00:45:02.580 --> 00:45:05.210 align:start position:0% farm can be reused is that the case that is<00:45:02.760> very<00:45:03.030> much<00:45:03.210> the<00:45:03.420> case<00:45:03.600> I<00:45:04.170> can<00:45:04.980> talk<00:45:05.190> a 00:45:05.210 --> 00:45:05.220 align:start position:0% is very much the case I can talk a 00:45:05.220 --> 00:45:07.070 align:start position:0% is very much the case I can talk a little<00:45:05.460> bit<00:45:05.640> about<00:45:05.820> how<00:45:06.000> awesome<00:45:06.210> GPUs<00:45:06.869> are 00:45:07.070 --> 00:45:07.080 align:start position:0% little bit about how awesome GPUs are 00:45:07.080 --> 00:45:09.890 align:start position:0% little bit about how awesome GPUs are they're<00:45:07.290> repurpose<00:45:07.890> 'fl<00:45:08.070> I<00:45:08.280> love<00:45:08.640> that<00:45:08.900> you 00:45:09.890 --> 00:45:09.900 align:start position:0% they're repurpose 'fl I love that you 00:45:09.900 --> 00:45:11.750 align:start position:0% they're repurpose 'fl I love that you can<00:45:10.020> play<00:45:10.200> a<00:45:10.230> game<00:45:10.320> on<00:45:10.680> it<00:45:10.830> though<00:45:11.040> most<00:45:11.400> miners 00:45:11.750 --> 00:45:11.760 align:start position:0% can play a game on it though most miners 00:45:11.760 --> 00:45:13.220 align:start position:0% can play a game on it though most miners won't<00:45:11.970> be<00:45:12.090> playing<00:45:12.330> games<00:45:12.570> with<00:45:12.750> their<00:45:12.900> mining 00:45:13.220 --> 00:45:13.230 align:start position:0% won't be playing games with their mining 00:45:13.230 --> 00:45:16.609 align:start position:0% won't be playing games with their mining farms<00:45:13.619> but<00:45:14.190> you<00:45:14.340> can<00:45:14.550> use<00:45:14.790> alternate<00:45:15.500> you<00:45:16.500> know 00:45:16.609 --> 00:45:16.619 align:start position:0% farms but you can use alternate you know 00:45:16.619 --> 00:45:20.300 align:start position:0% farms but you can use alternate you know alternate<00:45:17.119> proofs<00:45:18.119> now<00:45:18.420> a<00:45:18.690> golem<00:45:19.410> is<00:45:19.800> probably 00:45:20.300 --> 00:45:20.310 align:start position:0% alternate proofs now a golem is probably 00:45:20.310 --> 00:45:21.859 align:start position:0% alternate proofs now a golem is probably the<00:45:20.400> one<00:45:20.550> you're<00:45:20.670> most<00:45:20.850> familiar<00:45:21.150> with<00:45:21.630> you 00:45:21.859 --> 00:45:21.869 align:start position:0% the one you're most familiar with you 00:45:21.869 --> 00:45:23.450 align:start position:0% the one you're most familiar with you can<00:45:21.900> sell<00:45:22.170> your<00:45:22.290> GPU<00:45:22.590> cycles<00:45:23.070> now<00:45:23.220> for 00:45:23.450 --> 00:45:23.460 align:start position:0% can sell your GPU cycles now for 00:45:23.460 --> 00:45:25.700 align:start position:0% can sell your GPU cycles now for rendering<00:45:23.850> so<00:45:24.480> that<00:45:24.630> gives<00:45:24.840> mining<00:45:25.260> farms<00:45:25.590> a 00:45:25.700 --> 00:45:25.710 align:start position:0% rendering so that gives mining farms a 00:45:25.710 --> 00:45:27.380 align:start position:0% rendering so that gives mining farms a little<00:45:25.920> bit<00:45:26.040> more<00:45:26.250> of<00:45:26.400> a<00:45:26.490> profitability<00:45:27.000> and 00:45:27.380 --> 00:45:27.390 align:start position:0% little bit more of a profitability and 00:45:27.390 --> 00:45:29.740 align:start position:0% little bit more of a profitability and that<00:45:27.900> contributes<00:45:28.470> to<00:45:28.619> a<00:45:28.650> pretty<00:45:28.890> cool<00:45:29.070> world 00:45:29.740 --> 00:45:29.750 align:start position:0% that contributes to a pretty cool world 00:45:29.750 --> 00:45:32.180 align:start position:0% that contributes to a pretty cool world proof<00:45:30.750> of<00:45:30.900> useful<00:45:31.290> work<00:45:31.320> is<00:45:31.710> something<00:45:32.010> that's 00:45:32.180 --> 00:45:32.190 align:start position:0% proof of useful work is something that's 00:45:32.190 --> 00:45:34.250 align:start position:0% proof of useful work is something that's being<00:45:32.369> explored<00:45:32.730> as<00:45:33.060> well<00:45:33.210> trying<00:45:33.840> to<00:45:33.960> find<00:45:34.170> a 00:45:34.250 --> 00:45:34.260 align:start position:0% being explored as well trying to find a 00:45:34.260 --> 00:45:36.099 align:start position:0% being explored as well trying to find a use<00:45:34.500> case<00:45:34.710> for<00:45:34.980> all<00:45:35.070> these<00:45:35.280> proof-of-work 00:45:36.099 --> 00:45:36.109 align:start position:0% use case for all these proof-of-work 00:45:36.109 --> 00:45:40.099 align:start position:0% use case for all these proof-of-work machines<00:45:37.109> out<00:45:37.440> there<00:45:38.780> there<00:45:39.780> are<00:45:39.930> some 00:45:40.099 --> 00:45:40.109 align:start position:0% machines out there there are some 00:45:40.109 --> 00:45:41.870 align:start position:0% machines out there there are some interesting<00:45:40.410> things<00:45:40.650> with<00:45:41.100> computational 00:45:41.870 --> 00:45:41.880 align:start position:0% interesting things with computational 00:45:41.880 --> 00:45:44.599 align:start position:0% interesting things with computational fluid<00:45:42.410> dynamic<00:45:43.410> annex<00:45:43.770> there's<00:45:44.400> some 00:45:44.599 --> 00:45:44.609 align:start position:0% fluid dynamic annex there's some 00:45:44.609 --> 00:45:46.250 align:start position:0% fluid dynamic annex there's some interesting<00:45:45.060> stuff<00:45:45.330> in<00:45:45.450> the<00:45:45.540> medical<00:45:45.990> space 00:45:46.250 --> 00:45:46.260 align:start position:0% interesting stuff in the medical space 00:45:46.260 --> 00:45:48.980 align:start position:0% interesting stuff in the medical space nothing<00:45:47.040> is<00:45:47.340> being<00:45:47.520> fleshed<00:45:47.940> out<00:45:48.210> to<00:45:48.690> date 00:45:48.980 --> 00:45:48.990 align:start position:0% nothing is being fleshed out to date 00:45:48.990 --> 00:45:51.109 align:start position:0% nothing is being fleshed out to date simply<00:45:49.859> because<00:45:50.160> people<00:45:50.460> haven't<00:45:50.790> solved<00:45:51.060> the 00:45:51.109 --> 00:45:51.119 align:start position:0% simply because people haven't solved the 00:45:51.119 --> 00:45:52.910 align:start position:0% simply because people haven't solved the liability<00:45:51.810> problem<00:45:52.170> that<00:45:52.350> comes<00:45:52.560> with<00:45:52.770> having 00:45:52.910 --> 00:45:52.920 align:start position:0% liability problem that comes with having 00:45:52.920 --> 00:45:56.660 align:start position:0% liability problem that comes with having a<00:45:53.070> very<00:45:53.250> decentralized<00:45:54.560> system<00:45:55.560> but<00:45:56.190> golems 00:45:56.660 --> 00:45:56.670 align:start position:0% a very decentralized system but golems 00:45:56.670 --> 00:45:58.840 align:start position:0% a very decentralized system but golems on<00:45:56.850> the<00:45:56.940> right<00:45:57.090> track<00:45:57.390> which<00:45:58.020> is<00:45:58.170> great<00:45:58.410> and 00:45:58.840 --> 00:45:58.850 align:start position:0% on the right track which is great and 00:45:58.850 --> 00:46:02.090 align:start position:0% on the right track which is great and that's<00:45:59.850> taken<00:46:00.840> what<00:46:01.140> just<00:46:01.530> a<00:46:01.619> year<00:46:01.920> of 00:46:02.090 --> 00:46:02.100 align:start position:0% that's taken what just a year of 00:46:02.100 --> 00:46:03.680 align:start position:0% that's taken what just a year of development<00:46:02.700> imagine<00:46:03.060> what<00:46:03.180> we're<00:46:03.330> gonna<00:46:03.450> see 00:46:03.680 --> 00:46:03.690 align:start position:0% development imagine what we're gonna see 00:46:03.690 --> 00:46:06.260 align:start position:0% development imagine what we're gonna see next<00:46:03.869> year<00:46:04.050> so<00:46:05.040> yeah<00:46:05.250> GPUs<00:46:05.790> are<00:46:05.970> pretty 00:46:06.260 --> 00:46:06.270 align:start position:0% next year so yeah GPUs are pretty 00:46:06.270 --> 00:46:08.450 align:start position:0% next year so yeah GPUs are pretty awesome<00:46:06.450> because<00:46:06.810> they're<00:46:07.050> repurposed<00:46:07.619> and 00:46:08.450 --> 00:46:08.460 align:start position:0% awesome because they're repurposed and 00:46:08.460 --> 00:46:12.080 align:start position:0% awesome because they're repurposed and on<00:46:09.180> that<00:46:09.390> note<00:46:09.570> FPGAs<00:46:10.380> are<00:46:10.650> as<00:46:10.859> well<00:46:11.070> but<00:46:11.940> they 00:46:12.080 --> 00:46:12.090 align:start position:0% on that note FPGAs are as well but they 00:46:12.090 --> 00:46:14.870 align:start position:0% on that note FPGAs are as well but they do<00:46:12.150> have<00:46:12.570> a<00:46:12.600> much<00:46:13.020> higher<00:46:13.230> skill<00:46:13.680> level<00:46:13.980> you<00:46:14.190> do 00:46:14.870 --> 00:46:14.880 align:start position:0% do have a much higher skill level you do 00:46:14.880 --> 00:46:17.060 align:start position:0% do have a much higher skill level you do need<00:46:15.090> to<00:46:15.210> go<00:46:15.750> and<00:46:15.960> rewrite<00:46:16.440> the<00:46:16.590> bit<00:46:16.710> streams 00:46:17.060 --> 00:46:17.070 align:start position:0% need to go and rewrite the bit streams 00:46:17.070 --> 00:46:19.700 align:start position:0% need to go and rewrite the bit streams you<00:46:17.700> need<00:46:17.850> to<00:46:18.000> reconfigure<00:46:18.480> them<00:46:18.810> tune<00:46:19.410> them 00:46:19.700 --> 00:46:19.710 align:start position:0% you need to reconfigure them tune them 00:46:19.710 --> 00:46:22.130 align:start position:0% you need to reconfigure them tune them and<00:46:20.040> the<00:46:20.369> amount<00:46:20.640> of<00:46:20.730> GPU<00:46:21.210> compute<00:46:21.660> architects 00:46:22.130 --> 00:46:22.140 align:start position:0% and the amount of GPU compute architects 00:46:22.140 --> 00:46:25.099 align:start position:0% and the amount of GPU compute architects out<00:46:22.320> there<00:46:23.210> significantly<00:46:24.210> Dwarfs<00:46:24.930> the 00:46:25.099 --> 00:46:25.109 align:start position:0% out there significantly Dwarfs the 00:46:25.109 --> 00:46:30.140 align:start position:0% out there significantly Dwarfs the amount<00:46:25.440> of<00:46:25.590> VHDL<00:46:26.369> and<00:46:26.609> Verilog<00:46:27.080> architects<00:46:29.150> hi 00:46:30.140 --> 00:46:30.150 align:start position:0% amount of VHDL and Verilog architects hi 00:46:30.150 --> 00:46:33.200 align:start position:0% amount of VHDL and Verilog architects hi great<00:46:30.450> talk<00:46:30.890> I'm<00:46:31.890> just<00:46:32.130> wondering<00:46:32.580> you<00:46:33.180> have 00:46:33.200 --> 00:46:33.210 align:start position:0% great talk I'm just wondering you have 00:46:33.210 --> 00:46:35.870 align:start position:0% great talk I'm just wondering you have tunable<00:46:34.080> parameters<00:46:34.410> these<00:46:35.220> are<00:46:35.460> integers 00:46:35.870 --> 00:46:35.880 align:start position:0% tunable parameters these are integers 00:46:35.880 --> 00:46:37.040 align:start position:0% tunable parameters these are integers yep 00:46:37.040 --> 00:46:37.050 align:start position:0% yep 00:46:37.050 --> 00:46:40.430 align:start position:0% yep how<00:46:37.800> about<00:46:37.890> tuning<00:46:38.540> which<00:46:39.540> app<00:46:39.750> codes<00:46:40.020> we<00:46:40.230> use 00:46:40.430 --> 00:46:40.440 align:start position:0% how about tuning which app codes we use 00:46:40.440 --> 00:46:42.440 align:start position:0% how about tuning which app codes we use there<00:46:41.430> I<00:46:41.460> think<00:46:41.580> there<00:46:41.790> was<00:46:41.910> like<00:46:41.940> a<00:46:42.060> list<00:46:42.300> of 00:46:42.440 --> 00:46:42.450 align:start position:0% there I think there was like a list of 00:46:42.450 --> 00:46:44.450 align:start position:0% there I think there was like a list of 11<00:46:42.810> app<00:46:42.960> codes<00:46:43.230> that<00:46:43.380> we<00:46:43.500> choose<00:46:43.680> randomly<00:46:44.100> and 00:46:44.450 --> 00:46:44.460 align:start position:0% 11 app codes that we choose randomly and 00:46:44.460 --> 00:46:47.840 align:start position:0% 11 app codes that we choose randomly and maybe<00:46:45.330> tuning<00:46:45.780> also<00:46:46.320> the<00:46:46.890> hashing<00:46:47.280> algorithm 00:46:47.840 --> 00:46:47.850 align:start position:0% maybe tuning also the hashing algorithm 00:46:47.850 --> 00:46:50.210 align:start position:0% maybe tuning also the hashing algorithm maybe<00:46:48.660> something<00:46:49.050> that<00:46:49.080> you<00:46:49.470> know<00:46:49.890> you<00:46:49.950> say 00:46:50.210 --> 00:46:50.220 align:start position:0% maybe something that you know you say 00:46:50.220 --> 00:46:52.130 align:start position:0% maybe something that you know you say that<00:46:50.250> some<00:46:50.880> version<00:46:51.180> of<00:46:51.240> ketchup<00:46:51.660> is<00:46:51.840> is 00:46:52.130 --> 00:46:52.140 align:start position:0% that some version of ketchup is is 00:46:52.140 --> 00:46:53.930 align:start position:0% that some version of ketchup is is better<00:46:52.410> for<00:46:52.890> some<00:46:53.070> GPUs<00:46:53.460> how<00:46:53.640> about<00:46:53.760> some 00:46:53.930 --> 00:46:53.940 align:start position:0% better for some GPUs how about some 00:46:53.940 --> 00:46:56.120 align:start position:0% better for some GPUs how about some different<00:46:54.360> algorithm<00:46:54.810> altogether<00:46:55.230> or<00:46:55.860> maybe 00:46:56.120 --> 00:46:56.130 align:start position:0% different algorithm altogether or maybe 00:46:56.130 --> 00:46:57.890 align:start position:0% different algorithm altogether or maybe two<00:46:56.250> different<00:46:56.370> algorithms<00:46:57.000> one<00:46:57.420> different 00:46:57.890 --> 00:46:57.900 align:start position:0% two different algorithms one different 00:46:57.900 --> 00:46:59.170 align:start position:0% two different algorithms one different one<00:46:57.990> at<00:46:58.110> the<00:46:58.170> beginning<00:46:58.260> at<00:46:58.590> the<00:46:58.680> end<00:46:58.800> just 00:46:59.170 --> 00:46:59.180 align:start position:0% one at the beginning at the end just 00:46:59.180 --> 00:47:01.460 align:start position:0% one at the beginning at the end just someone<00:47:00.180> that's<00:47:00.330> naive<00:47:00.720> that<00:47:01.020> says<00:47:01.230> why<00:47:01.410> this 00:47:01.460 --> 00:47:01.470 align:start position:0% someone that's naive that says why this 00:47:01.470 --> 00:47:02.450 align:start position:0% someone that's naive that says why this why<00:47:01.920> that 00:47:02.450 --> 00:47:02.460 align:start position:0% why that 00:47:02.460 --> 00:47:05.000 align:start position:0% why that poking<00:47:03.360> holes<00:47:03.480> you<00:47:03.990> know<00:47:04.080> at<00:47:04.260> every<00:47:04.620> part<00:47:04.950> of 00:47:05.000 --> 00:47:05.010 align:start position:0% poking holes you know at every part of 00:47:05.010 --> 00:47:06.530 align:start position:0% poking holes you know at every part of your<00:47:05.130> diagram<00:47:05.640> can<00:47:06.120> we<00:47:06.240> do<00:47:06.360> something 00:47:06.530 --> 00:47:06.540 align:start position:0% your diagram can we do something 00:47:06.540 --> 00:47:09.350 align:start position:0% your diagram can we do something different<00:47:06.720> here<00:47:07.080> we<00:47:07.440> can<00:47:07.950> the<00:47:08.610> biggest<00:47:09.000> thing 00:47:09.350 --> 00:47:09.360 align:start position:0% different here we can the biggest thing 00:47:09.360 --> 00:47:11.450 align:start position:0% different here we can the biggest thing you<00:47:09.540> don't<00:47:09.750> want<00:47:10.020> to<00:47:10.080> do<00:47:10.230> though<00:47:10.410> is<00:47:10.620> one<00:47:11.190> we 00:47:11.450 --> 00:47:11.460 align:start position:0% you don't want to do though is one we 00:47:11.460 --> 00:47:13.370 align:start position:0% you don't want to do though is one we wanted<00:47:11.730> to<00:47:11.820> make<00:47:11.970> Prague<00:47:12.240> towers<00:47:12.630> simple<00:47:13.140> at 00:47:13.370 --> 00:47:13.380 align:start position:0% wanted to make Prague towers simple at 00:47:13.380 --> 00:47:16.430 align:start position:0% wanted to make Prague towers simple at today<00:47:13.950> to<00:47:14.610> slot<00:47:14.940> into<00:47:15.270> the<00:47:15.600> existing<00:47:16.110> at<00:47:16.230> hash 00:47:16.430 --> 00:47:16.440 align:start position:0% today to slot into the existing at hash 00:47:16.440 --> 00:47:18.710 align:start position:0% today to slot into the existing at hash implementation<00:47:17.250> so<00:47:18.060> we<00:47:18.180> didn't<00:47:18.390> want<00:47:18.510> to<00:47:18.600> get 00:47:18.710 --> 00:47:18.720 align:start position:0% implementation so we didn't want to get 00:47:18.720 --> 00:47:20.600 align:start position:0% implementation so we didn't want to get dick<00:47:19.140> around<00:47:19.290> too<00:47:19.680> much<00:47:19.890> because<00:47:20.160> that<00:47:20.460> would 00:47:20.600 --> 00:47:20.610 align:start position:0% dick around too much because that would 00:47:20.610 --> 00:47:22.550 align:start position:0% dick around too much because that would have<00:47:20.670> required<00:47:20.820> way<00:47:21.210> more<00:47:21.480> test<00:47:21.750> analysis<00:47:22.320> it 00:47:22.550 --> 00:47:22.560 align:start position:0% have required way more test analysis it 00:47:22.560 --> 00:47:24.950 align:start position:0% have required way more test analysis it dicks<00:47:23.190> around<00:47:23.550> with<00:47:23.700> the<00:47:23.820> CPU<00:47:24.150> verification 00:47:24.950 --> 00:47:24.960 align:start position:0% dicks around with the CPU verification 00:47:24.960 --> 00:47:27.290 align:start position:0% dicks around with the CPU verification as<00:47:25.110> well<00:47:25.260> which<00:47:25.950> we<00:47:26.490> know<00:47:26.640> is<00:47:26.790> very<00:47:26.850> very 00:47:27.290 --> 00:47:27.300 align:start position:0% as well which we know is very very 00:47:27.300 --> 00:47:29.900 align:start position:0% as well which we know is very very important<00:47:27.810> to<00:47:27.870> the<00:47:28.110> etherium<00:47:28.530> eco<00:47:28.800> eco<00:47:29.370> system 00:47:29.900 --> 00:47:29.910 align:start position:0% important to the etherium eco eco system 00:47:29.910 --> 00:47:32.780 align:start position:0% important to the etherium eco eco system and<00:47:30.860> quite<00:47:31.860> frankly<00:47:32.190> as<00:47:32.340> long<00:47:32.460> as<00:47:32.580> you<00:47:32.670> have 00:47:32.780 --> 00:47:32.790 align:start position:0% and quite frankly as long as you have 00:47:32.790 --> 00:47:34.670 align:start position:0% and quite frankly as long as you have enough<00:47:33.030> random<00:47:33.300> math<00:47:33.690> in<00:47:33.900> there<00:47:34.080> it's<00:47:34.350> going 00:47:34.670 --> 00:47:34.680 align:start position:0% enough random math in there it's going 00:47:34.680 --> 00:47:37.490 align:start position:0% enough random math in there it's going to<00:47:34.850> it's<00:47:35.850> going<00:47:36.000> to<00:47:36.450> make<00:47:36.690> it<00:47:36.870> impossible<00:47:37.170> to 00:47:37.490 --> 00:47:37.500 align:start position:0% to it's going to make it impossible to 00:47:37.500 --> 00:47:40.730 align:start position:0% to it's going to make it impossible to offload<00:47:38.030> we<00:47:39.030> could<00:47:39.270> replace<00:47:39.600> ket<00:47:40.020> Jack<00:47:40.290> with 00:47:40.730 --> 00:47:40.740 align:start position:0% offload we could replace ket Jack with 00:47:40.740 --> 00:47:42.920 align:start position:0% offload we could replace ket Jack with another<00:47:40.770> hash<00:47:41.400> function<00:47:41.880> I<00:47:42.090> haven't<00:47:42.720> really 00:47:42.920 --> 00:47:42.930 align:start position:0% another hash function I haven't really 00:47:42.930 --> 00:47:44.840 align:start position:0% another hash function I haven't really thought<00:47:42.960> too<00:47:43.560> much<00:47:43.590> about<00:47:44.010> it<00:47:44.220> we<00:47:44.430> also<00:47:44.580> had<00:47:44.820> a 00:47:44.840 --> 00:47:44.850 align:start position:0% thought too much about it we also had a 00:47:44.850 --> 00:47:46.820 align:start position:0% thought too much about it we also had a really<00:47:45.210> cool<00:47:45.330> idea<00:47:45.390> of<00:47:45.780> merge<00:47:46.080> mining<00:47:46.500> as<00:47:46.650> well 00:47:46.820 --> 00:47:46.830 align:start position:0% really cool idea of merge mining as well 00:47:46.830 --> 00:47:49.250 align:start position:0% really cool idea of merge mining as well which<00:47:47.760> would<00:47:47.940> be<00:47:48.060> a<00:47:48.090> great<00:47:48.270> way<00:47:48.570> where<00:47:48.900> you 00:47:49.250 --> 00:47:49.260 align:start position:0% which would be a great way where you 00:47:49.260 --> 00:47:50.870 align:start position:0% which would be a great way where you know<00:47:49.290> other<00:47:49.650> proof-of-work<00:47:50.130> algorithms 00:47:50.870 --> 00:47:50.880 align:start position:0% know other proof-of-work algorithms 00:47:50.880 --> 00:47:53.330 align:start position:0% know other proof-of-work algorithms could<00:47:51.150> use<00:47:51.330> probe<00:47:51.600> pal<00:47:51.810> in<00:47:52.110> tandem<00:47:52.860> to<00:47:53.010> compute 00:47:53.330 --> 00:47:53.340 align:start position:0% could use probe pal in tandem to compute 00:47:53.340 --> 00:47:54.860 align:start position:0% could use probe pal in tandem to compute their<00:47:53.460> final<00:47:53.820> cycles<00:47:54.240> to<00:47:54.420> get<00:47:54.510> a<00:47:54.570> bit<00:47:54.750> more 00:47:54.860 --> 00:47:54.870 align:start position:0% their final cycles to get a bit more 00:47:54.870 --> 00:47:56.870 align:start position:0% their final cycles to get a bit more ASIC<00:47:55.350> resistance<00:47:55.890> that<00:47:56.370> would<00:47:56.520> actually 00:47:56.870 --> 00:47:56.880 align:start position:0% ASIC resistance that would actually 00:47:56.880 --> 00:47:59.300 align:start position:0% ASIC resistance that would actually defeat<00:47:57.420> the<00:47:57.630> nice<00:47:57.870> hash<00:47:58.140> problem<00:47:58.590> that<00:47:59.100> we<00:47:59.190> all 00:47:59.300 --> 00:47:59.310 align:start position:0% defeat the nice hash problem that we all 00:47:59.310 --> 00:48:02.150 align:start position:0% defeat the nice hash problem that we all have<00:47:59.490> in<00:47:59.640> mining<00:48:00.590> but<00:48:01.590> that's<00:48:01.740> a<00:48:01.920> lot<00:48:02.100> of 00:48:02.150 --> 00:48:02.160 align:start position:0% have in mining but that's a lot of 00:48:02.160 --> 00:48:06.530 align:start position:0% have in mining but that's a lot of development<00:48:02.790> work<00:48:02.820> as<00:48:03.120> well<00:48:03.300> and<00:48:04.760> these<00:48:05.760> just 00:48:06.530 --> 00:48:06.540 align:start position:0% development work as well and these just 00:48:06.540 --> 00:48:08.570 align:start position:0% development work as well and these just to<00:48:06.660> be<00:48:06.750> clear<00:48:06.960> these<00:48:07.140> tunable<00:48:07.740> parameters<00:48:07.890> are 00:48:08.570 --> 00:48:08.580 align:start position:0% to be clear these tunable parameters are 00:48:08.580 --> 00:48:12.020 align:start position:0% to be clear these tunable parameters are just<00:48:09.000> for<00:48:09.390> future<00:48:10.170> revisions<00:48:10.770> of<00:48:10.890> GPUs<00:48:11.370> so 00:48:12.020 --> 00:48:12.030 align:start position:0% just for future revisions of GPUs so 00:48:12.030 --> 00:48:14.810 align:start position:0% just for future revisions of GPUs so today<00:48:12.360> all<00:48:12.690> of<00:48:12.990> these<00:48:13.170> are<00:48:13.590> completely<00:48:14.280> tuned 00:48:14.810 --> 00:48:14.820 align:start position:0% today all of these are completely tuned 00:48:14.820 --> 00:48:17.690 align:start position:0% today all of these are completely tuned for<00:48:15.180> NVIDIA<00:48:15.930> z--<00:48:16.140> latest<00:48:16.470> line<00:48:16.650> of<00:48:16.800> GPUs<00:48:17.190> which 00:48:17.690 --> 00:48:17.700 align:start position:0% for NVIDIA z-- latest line of GPUs which 00:48:17.700 --> 00:48:22.250 align:start position:0% for NVIDIA z-- latest line of GPUs which is<00:48:17.880> the<00:48:18.000> RT<00:48:18.330> X<00:48:18.480> to<00:48:19.110> 2000<00:48:20.210> a.m.<00:48:21.210> DS<00:48:21.420> latest<00:48:21.780> line 00:48:22.250 --> 00:48:22.260 align:start position:0% is the RT X to 2000 a.m. DS latest line 00:48:22.260 --> 00:48:24.980 align:start position:0% is the RT X to 2000 a.m. DS latest line which<00:48:22.530> is<00:48:22.680> their<00:48:22.830> Vega<00:48:23.240> version<00:48:24.240> 3<00:48:24.510> I<00:48:24.720> guess 00:48:24.980 --> 00:48:24.990 align:start position:0% which is their Vega version 3 I guess 00:48:24.990 --> 00:48:28.850 align:start position:0% which is their Vega version 3 I guess they're<00:48:25.170> on<00:48:25.200> now<00:48:25.470> and<00:48:26.150> also<00:48:27.290> we<00:48:28.290> did<00:48:28.500> have<00:48:28.740> an 00:48:28.850 --> 00:48:28.860 align:start position:0% they're on now and also we did have an 00:48:28.860 --> 00:48:31.310 align:start position:0% they're on now and also we did have an Intel<00:48:29.220> GPU<00:48:29.670> sample<00:48:30.210> but<00:48:30.420> I<00:48:30.450> don't<00:48:30.750> think<00:48:31.080> that 00:48:31.310 --> 00:48:31.320 align:start position:0% Intel GPU sample but I don't think that 00:48:31.320 --> 00:48:34.300 align:start position:0% Intel GPU sample but I don't think that those<00:48:31.680> will<00:48:31.950> be<00:48:32.100> massively<00:48:33.060> used<00:48:33.420> by<00:48:33.630> miners 00:48:34.300 --> 00:48:34.310 align:start position:0% those will be massively used by miners 00:48:34.310 --> 00:48:38.330 align:start position:0% those will be massively used by miners unless<00:48:35.310> you're<00:48:35.460> very<00:48:35.490> very<00:48:35.790> rich<00:48:36.650> so<00:48:37.650> a<00:48:37.680> great 00:48:38.330 --> 00:48:38.340 align:start position:0% unless you're very very rich so a great 00:48:38.340 --> 00:48:42.200 align:start position:0% unless you're very very rich so a great talk<00:48:38.610> thank<00:48:39.120> you<00:48:39.920> even<00:48:40.920> though<00:48:40.950> I<00:48:41.100> think<00:48:42.030> I 00:48:42.200 --> 00:48:42.210 align:start position:0% talk thank you even though I think I 00:48:42.210 --> 00:48:46.490 align:start position:0% talk thank you even though I think I probably<00:48:42.540> only<00:48:43.140> get<00:48:43.470> 10%<00:48:43.800> of<00:48:44.070> it<00:48:44.450> I<00:48:45.500> wonder 00:48:46.490 --> 00:48:46.500 align:start position:0% probably only get 10% of it I wonder 00:48:46.500 --> 00:48:50.120 align:start position:0% probably only get 10% of it I wonder from<00:48:47.720> integration<00:48:48.740> standpoint 00:48:50.120 --> 00:48:50.130 align:start position:0% from integration standpoint 00:48:50.130 --> 00:48:52.160 align:start position:0% from integration standpoint like<00:48:50.430> how<00:48:50.640> feasible<00:48:51.269> of<00:48:51.509> our<00:48:51.720> long<00:48:51.960> hour 00:48:52.160 --> 00:48:52.170 align:start position:0% like how feasible of our long hour 00:48:52.170 --> 00:48:55.609 align:start position:0% like how feasible of our long hour because<00:48:52.440> I<00:48:52.589> remember<00:48:53.250> I'm<00:48:53.519> seeing<00:48:54.059> that<00:48:54.990> the 00:48:55.609 --> 00:48:55.619 align:start position:0% because I remember I'm seeing that the 00:48:55.619 --> 00:48:59.599 align:start position:0% because I remember I'm seeing that the time<00:48:56.190> to<00:48:56.549> verify<00:48:57.089> the<00:48:57.150> proof<00:48:57.480> work<00:48:58.609> increased 00:48:59.599 --> 00:48:59.609 align:start position:0% time to verify the proof work increased 00:48:59.609 --> 00:49:02.120 align:start position:0% time to verify the proof work increased by<00:49:00.390> a<00:49:00.420> margin<00:49:00.779> that<00:49:01.230> might<00:49:01.769> make<00:49:02.099> it 00:49:02.120 --> 00:49:02.130 align:start position:0% by a margin that might make it 00:49:02.130 --> 00:49:06.039 align:start position:0% by a margin that might make it infeasible<00:49:03.029> to<00:49:03.089> actually<00:49:03.960> use<00:49:04.170> it<00:49:04.349> yeah<00:49:04.890> so 00:49:06.039 --> 00:49:06.049 align:start position:0% infeasible to actually use it yeah so 00:49:06.049 --> 00:49:08.779 align:start position:0% infeasible to actually use it yeah so that<00:49:07.049> let<00:49:07.559> let's<00:49:07.890> make<00:49:08.099> that<00:49:08.250> very<00:49:08.460> public 00:49:08.779 --> 00:49:08.789 align:start position:0% that let let's make that very public 00:49:08.789 --> 00:49:11.390 align:start position:0% that let let's make that very public that<00:49:09.539> is<00:49:09.809> simply<00:49:10.200> because<00:49:10.380> of<00:49:10.559> my<00:49:10.799> terrible<00:49:11.130> go 00:49:11.390 --> 00:49:11.400 align:start position:0% that is simply because of my terrible go 00:49:11.400 --> 00:49:16.009 align:start position:0% that is simply because of my terrible go skills<00:49:12.680> we<00:49:13.680> are<00:49:13.890> we<00:49:14.250> we<00:49:15.180> are<00:49:15.329> really<00:49:15.599> great<00:49:15.809> on 00:49:16.009 --> 00:49:16.019 align:start position:0% skills we are we we are really great on 00:49:16.019 --> 00:49:18.620 align:start position:0% skills we are we we are really great on the<00:49:16.200> GPU<00:49:16.500> side<00:49:16.799> not<00:49:17.160> so<00:49:17.190> great<00:49:17.579> on<00:49:17.819> you<00:49:18.450> know 00:49:18.620 --> 00:49:18.630 align:start position:0% the GPU side not so great on you know 00:49:18.630 --> 00:49:21.620 align:start position:0% the GPU side not so great on you know standard<00:49:19.319> client<00:49:19.769> code<00:49:19.980> we<00:49:20.789> fix<00:49:21.059> that<00:49:21.269> with<00:49:21.509> an 00:49:21.620 --> 00:49:21.630 align:start position:0% standard client code we fix that with an 00:49:21.630 --> 00:49:23.509 align:start position:0% standard client code we fix that with an optimized<00:49:22.049> implementation<00:49:22.289> had<00:49:23.190> some<00:49:23.339> great 00:49:23.509 --> 00:49:23.519 align:start position:0% optimized implementation had some great 00:49:23.519 --> 00:49:25.759 align:start position:0% optimized implementation had some great help<00:49:23.730> from<00:49:23.880> Paraty<00:49:24.569> devs<00:49:24.839> from<00:49:25.259> earth<00:49:25.500> minor 00:49:25.759 --> 00:49:25.769 align:start position:0% help from Paraty devs from earth minor 00:49:25.769 --> 00:49:27.920 align:start position:0% help from Paraty devs from earth minor developers<00:49:26.400> for<00:49:27.240> the<00:49:27.329> hope<00:49:27.450> from<00:49:27.660> the<00:49:27.809> whole 00:49:27.920 --> 00:49:27.930 align:start position:0% developers for the hope from the whole 00:49:27.930 --> 00:49:30.890 align:start position:0% developers for the hope from the whole ethereum<00:49:28.380> community<00:49:28.829> really<00:49:29.509> there<00:49:30.509> is<00:49:30.690> going 00:49:30.890 --> 00:49:30.900 align:start position:0% ethereum community really there is going 00:49:30.900 --> 00:49:32.870 align:start position:0% ethereum community really there is going to<00:49:31.109> be<00:49:31.230> a<00:49:31.470> -<00:49:31.829> it's<00:49:32.069> going<00:49:32.190> to<00:49:32.220> take<00:49:32.460> two<00:49:32.640> times 00:49:32.870 --> 00:49:32.880 align:start position:0% to be a - it's going to take two times 00:49:32.880 --> 00:49:35.539 align:start position:0% to be a - it's going to take two times as<00:49:33.329> longer<00:49:33.960> to<00:49:34.079> do<00:49:34.170> a<00:49:34.200> verification<00:49:34.549> simply 00:49:35.539 --> 00:49:35.549 align:start position:0% as longer to do a verification simply 00:49:35.549 --> 00:49:37.400 align:start position:0% as longer to do a verification simply because<00:49:35.819> two<00:49:36.000> times<00:49:36.240> the<00:49:36.450> data<00:49:36.630> is<00:49:36.869> consumed 00:49:37.400 --> 00:49:37.410 align:start position:0% because two times the data is consumed 00:49:37.410 --> 00:49:40.430 align:start position:0% because two times the data is consumed that's<00:49:37.589> the<00:49:37.769> 128<00:49:38.369> bytes<00:49:38.609> increase<00:49:39.569> to<00:49:39.750> 256 00:49:40.430 --> 00:49:40.440 align:start position:0% that's the 128 bytes increase to 256 00:49:40.440 --> 00:49:42.829 align:start position:0% that's the 128 bytes increase to 256 bytes<00:49:40.700> but<00:49:41.700> as<00:49:41.970> long<00:49:42.150> as<00:49:42.180> everyone's 00:49:42.829 --> 00:49:42.839 align:start position:0% bytes but as long as everyone's 00:49:42.839 --> 00:49:44.599 align:start position:0% bytes but as long as everyone's verifying<00:49:43.349> proof<00:49:43.619> of<00:49:43.769> work<00:49:43.950> the<00:49:44.130> same<00:49:44.339> it 00:49:44.599 --> 00:49:44.609 align:start position:0% verifying proof of work the same it 00:49:44.609 --> 00:49:47.809 align:start position:0% verifying proof of work the same it evens<00:49:45.509> out<00:49:45.660> and<00:49:45.900> it<00:49:45.990> levels<00:49:46.170> out<00:49:46.440> and<00:49:46.769> a<00:49:47.069> 2x<00:49:47.309> a 00:49:47.809 --> 00:49:47.819 align:start position:0% evens out and it levels out and a 2x a 00:49:47.819 --> 00:49:52.700 align:start position:0% evens out and it levels out and a 2x a 2x<00:49:48.210> slowdown<00:49:48.750> is<00:49:49.019> much<00:49:49.859> more<00:49:50.779> palatable<00:49:51.779> than 00:49:52.700 --> 00:49:52.710 align:start position:0% 2x slowdown is much more palatable than 00:49:52.710 --> 00:49:54.740 align:start position:0% 2x slowdown is much more palatable than what<00:49:52.980> it<00:49:53.069> originally<00:49:53.190> was<00:49:53.579> which<00:49:54.029> was<00:49:54.210> a<00:49:54.240> 9x 00:49:54.740 --> 00:49:54.750 align:start position:0% what it originally was which was a 9x 00:49:54.750 --> 00:49:57.620 align:start position:0% what it originally was which was a 9x slowdown<00:49:55.430> so<00:49:56.430> that's<00:49:56.849> what<00:49:57.029> happens<00:49:57.359> when<00:49:57.509> you 00:49:57.620 --> 00:49:57.630 align:start position:0% slowdown so that's what happens when you 00:49:57.630 --> 00:50:02.150 align:start position:0% slowdown so that's what happens when you don't<00:49:57.750> optimize<00:49:58.230> your<00:49:58.380> code<00:49:58.559> guys<00:50:00.559> hello<00:50:01.559> me 00:50:02.150 --> 00:50:02.160 align:start position:0% don't optimize your code guys hello me 00:50:02.160 --> 00:50:06.529 align:start position:0% don't optimize your code guys hello me again<00:50:03.049> on<00:50:04.049> the<00:50:04.410> legal<00:50:04.740> front<00:50:05.329> someone<00:50:06.329> might 00:50:06.529 --> 00:50:06.539 align:start position:0% again on the legal front someone might 00:50:06.539 --> 00:50:08.359 align:start position:0% again on the legal front someone might say<00:50:06.779> some<00:50:07.200> things<00:50:07.380> are<00:50:07.500> copyrighted<00:50:08.009> some 00:50:08.359 --> 00:50:08.369 align:start position:0% say some things are copyrighted some 00:50:08.369 --> 00:50:11.599 align:start position:0% say some things are copyrighted some code<00:50:08.730> is<00:50:08.940> sort<00:50:09.210> of<00:50:09.589> you<00:50:10.589> know<00:50:10.619> intellectual 00:50:11.599 --> 00:50:11.609 align:start position:0% code is sort of you know intellectual 00:50:11.609 --> 00:50:13.339 align:start position:0% code is sort of you know intellectual property<00:50:12.059> and<00:50:12.210> your<00:50:12.539> other<00:50:12.930> people<00:50:13.259> are 00:50:13.339 --> 00:50:13.349 align:start position:0% property and your other people are 00:50:13.349 --> 00:50:15.650 align:start position:0% property and your other people are running<00:50:13.500> this<00:50:13.740> code<00:50:14.150> would<00:50:15.150> there<00:50:15.299> be<00:50:15.329> any 00:50:15.650 --> 00:50:15.660 align:start position:0% running this code would there be any 00:50:15.660 --> 00:50:17.180 align:start position:0% running this code would there be any barrier<00:50:16.079> does<00:50:16.589> someone<00:50:16.829> own<00:50:17.009> this 00:50:17.180 --> 00:50:17.190 align:start position:0% barrier does someone own this 00:50:17.190 --> 00:50:18.769 align:start position:0% barrier does someone own this intellectual<00:50:17.730> property<00:50:18.059> does<00:50:18.180> anyone<00:50:18.630> own 00:50:18.769 --> 00:50:18.779 align:start position:0% intellectual property does anyone own 00:50:18.779 --> 00:50:21.589 align:start position:0% intellectual property does anyone own the<00:50:18.990> copyrights<00:50:19.410> to<00:50:19.500> Prague<00:50:19.799> POW<00:50:20.180> would<00:50:21.180> would 00:50:21.589 --> 00:50:21.599 align:start position:0% the copyrights to Prague POW would would 00:50:21.599 --> 00:50:23.329 align:start position:0% the copyrights to Prague POW would would there<00:50:21.750> be<00:50:21.779> someone<00:50:22.109> that<00:50:22.140> could<00:50:22.470> later<00:50:22.920> on<00:50:23.099> say 00:50:23.329 --> 00:50:23.339 align:start position:0% there be someone that could later on say 00:50:23.339 --> 00:50:25.789 align:start position:0% there be someone that could later on say okay<00:50:23.759> you're<00:50:24.000> using<00:50:24.210> my<00:50:24.450> code<00:50:25.170> you're<00:50:25.529> running 00:50:25.789 --> 00:50:25.799 align:start position:0% okay you're using my code you're running 00:50:25.799 --> 00:50:28.339 align:start position:0% okay you're using my code you're running it<00:50:25.950> now<00:50:26.190> you<00:50:26.490> know<00:50:26.519> I<00:50:26.819> mean<00:50:27.150> it's<00:50:27.480> open<00:50:28.049> source 00:50:28.339 --> 00:50:28.349 align:start position:0% it now you know I mean it's open source 00:50:28.349 --> 00:50:30.410 align:start position:0% it now you know I mean it's open source so<00:50:28.859> if<00:50:29.009> you<00:50:29.160> want<00:50:29.339> to<00:50:29.430> go<00:50:29.549> patent<00:50:29.970> the<00:50:30.089> open 00:50:30.410 --> 00:50:30.420 align:start position:0% so if you want to go patent the open 00:50:30.420 --> 00:50:32.630 align:start position:0% so if you want to go patent the open source<00:50:30.599> code<00:50:31.069> licensed<00:50:32.069> under<00:50:32.309> the<00:50:32.400> MIT 00:50:32.630 --> 00:50:32.640 align:start position:0% source code licensed under the MIT 00:50:32.640 --> 00:50:34.309 align:start position:0% source code licensed under the MIT license<00:50:33.269> go<00:50:33.480> for<00:50:33.720> it<00:50:33.809> guys 00:50:34.309 --> 00:50:34.319 align:start position:0% license go for it guys 00:50:34.319 --> 00:50:36.859 align:start position:0% license go for it guys no<00:50:35.009> there<00:50:35.220> is<00:50:35.309> no<00:50:35.430> copyright<00:50:35.670> again<00:50:36.509> this<00:50:36.690> has 00:50:36.859 --> 00:50:36.869 align:start position:0% no there is no copyright again this has 00:50:36.869 --> 00:50:39.680 align:start position:0% no there is no copyright again this has been<00:50:37.049> public<00:50:37.920> fully<00:50:38.400> public<00:50:38.849> fully<00:50:39.390> open 00:50:39.680 --> 00:50:39.690 align:start position:0% been public fully public fully open 00:50:39.690 --> 00:50:42.529 align:start position:0% been public fully public fully open source<00:50:39.950> since<00:50:40.950> I<00:50:41.369> think<00:50:41.789> we<00:50:41.910> released<00:50:42.269> it<00:50:42.420> in 00:50:42.529 --> 00:50:42.539 align:start position:0% source since I think we released it in 00:50:42.539 --> 00:50:49.810 align:start position:0% source since I think we released it in April 00:50:49.810 --> 00:50:49.820 align:start position:0% 00:50:49.820 --> 00:50:54.280 align:start position:0% oh<00:50:51.610> god<00:50:52.610> it's<00:50:52.880> GPL<00:50:53.300> infected<00:50:53.810> well<00:50:54.020> there<00:50:54.170> we 00:50:54.280 --> 00:50:54.290 align:start position:0% oh god it's GPL infected well there we 00:50:54.290 --> 00:50:54.660 align:start position:0% oh god it's GPL infected well there we go 00:50:54.660 --> 00:50:54.670 align:start position:0% go 00:50:54.670 --> 00:51:00.040 align:start position:0% go there<00:50:55.670> we<00:50:55.820> go 00:51:00.040 --> 00:51:00.050 align:start position:0% 00:51:00.050 --> 00:51:05.560 align:start position:0% I<00:51:00.080> have<00:51:00.680> several<00:51:01.160> questions<00:51:04.120> regarding<00:51:05.120> the 00:51:05.560 --> 00:51:05.570 align:start position:0% I have several questions regarding the 00:51:05.570 --> 00:51:09.450 align:start position:0% I have several questions regarding the specs<00:51:05.930> other<00:51:06.790> official<00:51:07.790> specs<00:51:08.090> now<00:51:08.450> because 00:51:09.450 --> 00:51:09.460 align:start position:0% specs other official specs now because 00:51:09.460 --> 00:51:17.640 align:start position:0% specs other official specs now because in<00:51:10.460> the<00:51:10.940> beginning<00:51:11.090> it<00:51:11.480> was<00:51:11.630> a<00:51:11.660> bit<00:51:12.730> janky<00:51:16.630> I 00:51:17.640 --> 00:51:17.650 align:start position:0% in the beginning it was a bit janky I 00:51:17.650 --> 00:51:20.950 align:start position:0% in the beginning it was a bit janky I realized<00:51:18.650> that<00:51:18.830> there<00:51:19.040> is<00:51:19.190> a<00:51:19.450> still<00:51:20.450> some<00:51:20.750> use 00:51:20.950 --> 00:51:20.960 align:start position:0% realized that there is a still some use 00:51:20.960 --> 00:51:27.340 align:start position:0% realized that there is a still some use of<00:51:21.140> FN<00:51:21.470> v<00:51:21.770> in<00:51:24.490> hey<00:51:25.490> no<00:51:26.270> no<00:51:26.570> the<00:51:26.750> normal<00:51:27.140> Wow<00:51:27.260> oh 00:51:27.340 --> 00:51:27.350 align:start position:0% of FN v in hey no no the normal Wow oh 00:51:27.350 --> 00:51:31.570 align:start position:0% of FN v in hey no no the normal Wow oh yeah<00:51:27.590> yeah<00:51:27.740> because<00:51:28.130> I've<00:51:28.970> I've<00:51:29.780> seen<00:51:30.290> the<00:51:30.580> the 00:51:31.570 --> 00:51:31.580 align:start position:0% yeah yeah because I've I've seen the the 00:51:31.580 --> 00:51:35.140 align:start position:0% yeah yeah because I've I've seen the the primes<00:51:31.910> used<00:51:32.450> by<00:51:32.660> the<00:51:32.720> normal<00:51:33.170> one<00:51:34.090> why<00:51:35.090> is 00:51:35.140 --> 00:51:35.150 align:start position:0% primes used by the normal one why is 00:51:35.150 --> 00:51:38.050 align:start position:0% primes used by the normal one why is that<00:51:35.920> simply<00:51:36.920> because<00:51:37.280> we<00:51:37.550> didn't<00:51:37.880> want<00:51:38.000> to 00:51:38.050 --> 00:51:38.060 align:start position:0% that simply because we didn't want to 00:51:38.060 --> 00:51:39.820 align:start position:0% that simply because we didn't want to mess<00:51:38.330> around<00:51:38.510> with<00:51:38.840> the<00:51:39.050> original<00:51:39.470> proof<00:51:39.710> of 00:51:39.820 --> 00:51:39.830 align:start position:0% mess around with the original proof of 00:51:39.830 --> 00:51:41.440 align:start position:0% mess around with the original proof of algorithm<00:51:40.310> too<00:51:40.460> much<00:51:40.670> because<00:51:40.940> it<00:51:41.180> affects 00:51:41.440 --> 00:51:41.450 align:start position:0% algorithm too much because it affects 00:51:41.450 --> 00:51:43.840 align:start position:0% algorithm too much because it affects that<00:51:41.630> you<00:51:41.810> it<00:51:42.200> affects<00:51:42.530> the<00:51:42.770> balance<00:51:43.160> of<00:51:43.670> the 00:51:43.840 --> 00:51:43.850 align:start position:0% that you it affects the balance of the 00:51:43.850 --> 00:51:48.820 align:start position:0% that you it affects the balance of the compute<00:51:44.480> core<00:51:44.660> of<00:51:44.780> the<00:51:44.900> GPU<00:51:45.790> if<00:51:47.620> it's<00:51:48.620> in<00:51:48.710> the 00:51:48.820 --> 00:51:48.830 align:start position:0% compute core of the GPU if it's in the 00:51:48.830 --> 00:51:53.050 align:start position:0% compute core of the GPU if it's in the is<00:51:49.010> it<00:51:49.190> in<00:51:49.310> the<00:51:49.400> dead<00:51:50.110> duck<00:51:51.110> generation<00:51:51.880> yes<00:51:52.880> it 00:51:53.050 --> 00:51:53.060 align:start position:0% is it in the dead duck generation yes it 00:51:53.060 --> 00:51:59.410 align:start position:0% is it in the dead duck generation yes it is<00:51:53.620> okay<00:51:57.280> then<00:51:58.280> the<00:51:58.460> question<00:51:58.880> is<00:51:59.060> if<00:51:59.270> we 00:51:59.410 --> 00:51:59.420 align:start position:0% is okay then the question is if we 00:51:59.420 --> 00:52:01.780 align:start position:0% is okay then the question is if we shouldn't<00:51:59.780> just<00:52:00.350> change<00:52:00.800> this<00:52:01.070> if<00:52:01.310> we<00:52:01.460> change 00:52:01.780 --> 00:52:01.790 align:start position:0% shouldn't just change this if we change 00:52:01.790 --> 00:52:04.990 align:start position:0% shouldn't just change this if we change prop<00:52:02.360> oh<00:52:02.600> we<00:52:03.290> should<00:52:03.530> maybe<00:52:03.890> just<00:52:04.250> change<00:52:04.700> this 00:52:04.990 --> 00:52:05.000 align:start position:0% prop oh we should maybe just change this 00:52:05.000 --> 00:52:07.810 align:start position:0% prop oh we should maybe just change this if<00:52:05.300> this<00:52:05.600> we<00:52:06.170> can<00:52:06.410> buy<00:52:06.560> from 00:52:07.810 --> 00:52:07.820 align:start position:0% if this we can buy from 00:52:07.820 --> 00:52:10.450 align:start position:0% if this we can buy from we<00:52:08.120> can<00:52:08.330> we<00:52:08.570> can<00:52:08.750> definitely<00:52:08.930> take<00:52:09.650> FNV<00:52:10.280> out 00:52:10.450 --> 00:52:10.460 align:start position:0% we can we can definitely take FNV out 00:52:10.460 --> 00:52:12.940 align:start position:0% we can we can definitely take FNV out altogether<00:52:11.170> we'd<00:52:12.170> need<00:52:12.380> to<00:52:12.530> find<00:52:12.800> another 00:52:12.940 --> 00:52:12.950 align:start position:0% altogether we'd need to find another 00:52:12.950 --> 00:52:16.570 align:start position:0% altogether we'd need to find another really<00:52:13.760> light<00:52:14.090> verification<00:52:14.930> fast<00:52:15.200> hash<00:52:15.580> that 00:52:16.570 --> 00:52:16.580 align:start position:0% really light verification fast hash that 00:52:16.580 --> 00:52:18.070 align:start position:0% really light verification fast hash that would<00:52:16.700> just<00:52:16.880> require<00:52:17.120> some<00:52:17.390> sort<00:52:17.630> of<00:52:17.750> research 00:52:18.070 --> 00:52:18.080 align:start position:0% would just require some sort of research 00:52:18.080 --> 00:52:20.460 align:start position:0% would just require some sort of research maybe<00:52:18.500> we<00:52:18.710> should<00:52:18.890> just<00:52:19.130> change<00:52:19.310> it<00:52:19.520> to<00:52:19.550> FN<00:52:19.940> v 00:52:20.460 --> 00:52:20.470 align:start position:0% maybe we should just change it to FN v 00:52:20.470 --> 00:52:25.800 align:start position:0% maybe we should just change it to FN v f11<00:52:21.470> a<00:52:21.620> yeah<00:52:22.190> we<00:52:22.490> can<00:52:22.640> do<00:52:22.760> that<00:52:22.940> maybe<00:52:23.480> yeah<00:52:23.840> ok 00:52:25.800 --> 00:52:25.810 align:start position:0% f11 a yeah we can do that maybe yeah ok 00:52:25.810 --> 00:52:33.040 align:start position:0% f11 a yeah we can do that maybe yeah ok then<00:52:26.830> the<00:52:27.830> cache<00:52:28.490> the<00:52:29.150> small<00:52:29.930> cache<00:52:30.380> in<00:52:32.050> shared 00:52:33.040 --> 00:52:33.050 align:start position:0% then the cache the small cache in shared 00:52:33.050 --> 00:52:35.350 align:start position:0% then the cache the small cache in shared memory<00:52:33.110> why<00:52:33.830> is<00:52:33.890> it<00:52:34.160> felt<00:52:34.460> that<00:52:34.640> way<00:52:34.820> it's<00:52:35.060> felt 00:52:35.350 --> 00:52:35.360 align:start position:0% memory why is it felt that way it's felt 00:52:35.360 --> 00:52:40.420 align:start position:0% memory why is it felt that way it's felt like<00:52:35.630> at<00:52:36.080> first<00:52:37.180> 128<00:52:38.180> bytes<00:52:39.110> from<00:52:39.710> from<00:52:40.310> the 00:52:40.420 --> 00:52:40.430 align:start position:0% like at first 128 bytes from from the 00:52:40.430 --> 00:52:42.520 align:start position:0% like at first 128 bytes from from the deck<00:52:40.610> or<00:52:40.850> something<00:52:41.210> or<00:52:41.360> the<00:52:41.510> first<00:52:42.020> elements 00:52:42.520 --> 00:52:42.530 align:start position:0% deck or something or the first elements 00:52:42.530 --> 00:52:46.540 align:start position:0% deck or something or the first elements from<00:52:42.650> the<00:52:42.830> deck<00:52:43.660> why<00:52:44.660> that<00:52:45.620> why<00:52:45.950> not<00:52:46.070> some 00:52:46.540 --> 00:52:46.550 align:start position:0% from the deck why that why not some 00:52:46.550 --> 00:52:49.600 align:start position:0% from the deck why that why not some randomness<00:52:47.060> or<00:52:47.510> something<00:52:48.280> simply<00:52:49.280> because 00:52:49.600 --> 00:52:49.610 align:start position:0% randomness or something simply because 00:52:49.610 --> 00:52:52.570 align:start position:0% randomness or something simply because that<00:52:49.850> would<00:52:50.120> so<00:52:50.900> if<00:52:51.440> you<00:52:51.770> were<00:52:51.920> going<00:52:52.220> to<00:52:52.340> make 00:52:52.570 --> 00:52:52.580 align:start position:0% that would so if you were going to make 00:52:52.580 --> 00:52:55.150 align:start position:0% that would so if you were going to make a<00:52:52.910> probe<00:52:53.360> pal<00:52:53.600> ASIC<00:52:54.110> because<00:52:54.740> it<00:52:54.860> doesn't 00:52:55.150 --> 00:52:55.160 align:start position:0% a probe pal ASIC because it doesn't 00:52:55.160 --> 00:52:58.870 align:start position:0% a probe pal ASIC because it doesn't saturate<00:52:55.370> all<00:52:56.210> of<00:52:56.600> it<00:52:56.750> completely<00:52:57.530> in<00:52:57.880> theory 00:52:58.870 --> 00:52:58.880 align:start position:0% saturate all of it completely in theory 00:52:58.880 --> 00:53:02.680 align:start position:0% saturate all of it completely in theory in<00:52:59.410> theory<00:53:00.410> you<00:53:01.220> could<00:53:01.370> go<00:53:01.550> and<00:53:01.700> create<00:53:02.000> a<00:53:02.210> sort 00:53:02.680 --> 00:53:02.690 align:start position:0% in theory you could go and create a sort 00:53:02.690 --> 00:53:03.730 align:start position:0% in theory you could go and create a sort of<00:53:02.810> a<00:53:02.900> compute<00:53:03.290> core 00:53:03.730 --> 00:53:03.740 align:start position:0% of a compute core 00:53:03.740 --> 00:53:06.940 align:start position:0% of a compute core that<00:53:04.520> takes<00:53:05.119> takes<00:53:05.720> all<00:53:05.930> this<00:53:06.080> randomness<00:53:06.530> and 00:53:06.940 --> 00:53:06.950 align:start position:0% that takes takes all this randomness and 00:53:06.950 --> 00:53:12.550 align:start position:0% that takes takes all this randomness and does<00:53:07.550> a<00:53:08.470> pre-compute<00:53:10.150> with<00:53:11.150> an<00:53:11.300> FPGA<00:53:11.780> it<00:53:12.380> would 00:53:12.550 --> 00:53:12.560 align:start position:0% does a pre-compute with an FPGA it would 00:53:12.560 --> 00:53:14.109 align:start position:0% does a pre-compute with an FPGA it would only<00:53:12.830> have<00:53:12.980> to<00:53:13.070> do<00:53:13.160> it<00:53:13.280> every<00:53:13.460> 12<00:53:13.670> minutes<00:53:14.030> you 00:53:14.109 --> 00:53:14.119 align:start position:0% only have to do it every 12 minutes you 00:53:14.119 --> 00:53:15.579 align:start position:0% only have to do it every 12 minutes you could<00:53:14.240> have<00:53:14.330> a<00:53:14.420> very<00:53:14.450> dedicated<00:53:14.839> bitstream 00:53:15.579 --> 00:53:15.589 align:start position:0% could have a very dedicated bitstream 00:53:15.589 --> 00:53:19.390 align:start position:0% could have a very dedicated bitstream developer<00:53:16.099> and<00:53:17.380> you<00:53:18.380> wouldn't<00:53:18.770> always<00:53:18.950> get 00:53:19.390 --> 00:53:19.400 align:start position:0% developer and you wouldn't always get 00:53:19.400 --> 00:53:22.420 align:start position:0% developer and you wouldn't always get that<00:53:19.640> balance<00:53:20.080> between<00:53:21.220> this<00:53:22.220> is<00:53:22.400> the 00:53:22.420 --> 00:53:22.430 align:start position:0% that balance between this is the 00:53:22.430 --> 00:53:23.740 align:start position:0% that balance between this is the important<00:53:22.940> part<00:53:23.060> you<00:53:23.240> wouldn't<00:53:23.510> get<00:53:23.599> the 00:53:23.740 --> 00:53:23.750 align:start position:0% important part you wouldn't get the 00:53:23.750 --> 00:53:25.690 align:start position:0% important part you wouldn't get the balance<00:53:24.080> between<00:53:24.230> a<00:53:24.440> NVIDIA<00:53:24.859> GPU<00:53:25.250> and<00:53:25.369> an<00:53:25.460> AMD 00:53:25.690 --> 00:53:25.700 align:start position:0% balance between a NVIDIA GPU and an AMD 00:53:25.700 --> 00:53:28.570 align:start position:0% balance between a NVIDIA GPU and an AMD GPU<00:53:26.710> sometimes<00:53:27.710> if<00:53:28.010> there<00:53:28.160> was<00:53:28.280> too<00:53:28.460> much 00:53:28.570 --> 00:53:28.580 align:start position:0% GPU sometimes if there was too much 00:53:28.580 --> 00:53:31.030 align:start position:0% GPU sometimes if there was too much randomness<00:53:29.150> you'd<00:53:29.390> end<00:53:29.599> up<00:53:29.720> going<00:53:30.109> way<00:53:30.619> over 00:53:31.030 --> 00:53:31.040 align:start position:0% randomness you'd end up going way over 00:53:31.040 --> 00:53:35.200 align:start position:0% randomness you'd end up going way over the<00:53:31.270> the<00:53:32.270> case<00:53:33.109> case<00:53:33.740> availability<00:53:34.490> for<00:53:35.089> an 00:53:35.200 --> 00:53:35.210 align:start position:0% the the case case availability for an 00:53:35.210 --> 00:53:38.020 align:start position:0% the the case case availability for an AMD<00:53:35.480> GPU<00:53:36.050> the<00:53:36.410> AMD<00:53:36.800> GPU<00:53:37.250> here<00:53:37.520> is<00:53:37.640> the<00:53:37.790> big 00:53:38.020 --> 00:53:38.030 align:start position:0% AMD GPU the AMD GPU here is the big 00:53:38.030 --> 00:53:41.940 align:start position:0% AMD GPU the AMD GPU here is the big bottleneck<00:53:38.950> and<00:53:39.950> I<00:53:40.369> think<00:53:41.060> both<00:53:41.270> both 00:53:41.940 --> 00:53:41.950 align:start position:0% bottleneck and I think both both 00:53:41.950 --> 00:53:44.160 align:start position:0% bottleneck and I think both both architectures<00:53:42.950> have<00:53:42.980> to<00:53:43.339> save<00:53:43.550> the<00:53:43.790> same 00:53:44.160 --> 00:53:44.170 align:start position:0% architectures have to save the same 00:53:44.170 --> 00:53:48.040 align:start position:0% architectures have to save the same amount<00:53:45.170> of<00:53:45.380> data<00:53:46.190> in<00:53:46.520> correct<00:53:47.300> but<00:53:47.810> if<00:53:47.900> you 00:53:48.040 --> 00:53:48.050 align:start position:0% amount of data in correct but if you 00:53:48.050 --> 00:53:50.380 align:start position:0% amount of data in correct but if you have<00:53:48.230> the<00:53:48.380> ran<00:53:48.560> randomness<00:53:49.430> changing<00:53:50.300> the 00:53:50.380 --> 00:53:50.390 align:start position:0% have the ran randomness changing the 00:53:50.390 --> 00:53:52.750 align:start position:0% have the ran randomness changing the bytes<00:53:50.599> that<00:53:51.050> screws<00:53:51.980> with<00:53:52.220> the<00:53:52.310> saturation 00:53:52.750 --> 00:53:52.760 align:start position:0% bytes that screws with the saturation 00:53:52.760 --> 00:53:55.960 align:start position:0% bytes that screws with the saturation and<00:53:53.150> then<00:53:54.080> that<00:53:54.290> would<00:53:54.500> cause<00:53:54.740> the<00:53:55.040> AMD<00:53:55.460> GPUs 00:53:55.960 --> 00:53:55.970 align:start position:0% and then that would cause the AMD GPUs 00:53:55.970 --> 00:53:58.210 align:start position:0% and then that would cause the AMD GPUs to<00:53:56.000> underperform<00:53:56.540> we<00:53:56.990> did<00:53:57.200> try<00:53:57.470> that<00:53:57.680> early<00:53:58.190> on 00:53:58.210 --> 00:53:58.220 align:start position:0% to underperform we did try that early on 00:53:58.220 --> 00:54:02.940 align:start position:0% to underperform we did try that early on it<00:53:58.700> didn't<00:53:59.000> work<00:53:59.119> out<00:53:59.270> too<00:53:59.330> well 00:54:02.940 --> 00:54:02.950 align:start position:0% 00:54:02.950 --> 00:54:07.020 align:start position:0% the<00:54:04.330> updates<00:54:05.330> so<00:54:05.570> we<00:54:05.750> have<00:54:05.930> an<00:54:06.080> update<00:54:06.410> every 00:54:07.020 --> 00:54:07.030 align:start position:0% the updates so we have an update every 00:54:07.030 --> 00:54:10.990 align:start position:0% the updates so we have an update every 25<00:54:08.030> I<00:54:08.270> don't<00:54:08.570> know<00:54:08.660> every<00:54:09.020> 50<00:54:09.500> or<00:54:09.800> 60<00:54:10.310> yeah<00:54:10.490> we 00:54:10.990 --> 00:54:11.000 align:start position:0% 25 I don't know every 50 or 60 yeah we 00:54:11.000 --> 00:54:13.960 align:start position:0% 25 I don't know every 50 or 60 yeah we put<00:54:11.210> it<00:54:11.300> at<00:54:11.420> 50<00:54:11.780> blocks<00:54:12.020> I<00:54:12.380> think<00:54:12.650> we<00:54:13.430> were<00:54:13.580> we 00:54:13.960 --> 00:54:13.970 align:start position:0% put it at 50 blocks I think we were we 00:54:13.970 --> 00:54:15.250 align:start position:0% put it at 50 blocks I think we were we finalized<00:54:14.450> on<00:54:14.599> 50<00:54:14.900> bucks 00:54:15.250 --> 00:54:15.260 align:start position:0% finalized on 50 bucks 00:54:15.260 --> 00:54:18.190 align:start position:0% finalized on 50 bucks you<00:54:15.950> could<00:54:16.250> change<00:54:16.580> it<00:54:16.849> to<00:54:16.910> 25<00:54:17.420> blocks<00:54:17.869> but 00:54:18.190 --> 00:54:18.200 align:start position:0% you could change it to 25 blocks but 00:54:18.200 --> 00:54:20.320 align:start position:0% you could change it to 25 blocks but there's<00:54:18.349> really<00:54:18.619> no<00:54:18.830> point<00:54:19.099> 50<00:54:19.880> blocks<00:54:20.150> is 00:54:20.320 --> 00:54:20.330 align:start position:0% there's really no point 50 blocks is 00:54:20.330 --> 00:54:22.150 align:start position:0% there's really no point 50 blocks is pretty<00:54:20.570> good<00:54:20.750> but<00:54:21.020> I<00:54:21.290> guess<00:54:21.619> if<00:54:21.890> you<00:54:21.950> want<00:54:22.099> to 00:54:22.150 --> 00:54:22.160 align:start position:0% pretty good but I guess if you want to 00:54:22.160 --> 00:54:24.460 align:start position:0% pretty good but I guess if you want to be<00:54:22.250> super<00:54:22.400> hardcore<00:54:22.790> changing<00:54:23.390> 25<00:54:23.869> blocks 00:54:24.460 --> 00:54:24.470 align:start position:0% be super hardcore changing 25 blocks 00:54:24.470 --> 00:54:27.070 align:start position:0% be super hardcore changing 25 blocks what's<00:54:24.950> the<00:54:25.130> overhead<00:54:25.520> of<00:54:25.700> this<00:54:25.910> update<00:54:26.119> do 00:54:27.070 --> 00:54:27.080 align:start position:0% what's the overhead of this update do 00:54:27.080 --> 00:54:29.170 align:start position:0% what's the overhead of this update do you<00:54:27.170> have<00:54:27.410> some<00:54:27.710> what<00:54:28.670> do<00:54:28.790> you<00:54:28.880> mean<00:54:29.030> by 00:54:29.170 --> 00:54:29.180 align:start position:0% you have some what do you mean by 00:54:29.180 --> 00:54:32.140 align:start position:0% you have some what do you mean by overhead<00:54:29.630> like<00:54:30.349> do<00:54:31.220> you<00:54:31.580> have<00:54:31.609> to<00:54:31.820> pre-compute 00:54:32.140 --> 00:54:32.150 align:start position:0% overhead like do you have to pre-compute 00:54:32.150 --> 00:54:35.500 align:start position:0% overhead like do you have to pre-compute the<00:54:32.480> kernel<00:54:32.869> do<00:54:33.770> on<00:54:34.070> a<00:54:34.130> GPU<00:54:34.609> it's<00:54:34.940> it's<00:54:35.270> really 00:54:35.500 --> 00:54:35.510 align:start position:0% the kernel do on a GPU it's it's really 00:54:35.510 --> 00:54:38.200 align:start position:0% the kernel do on a GPU it's it's really fast<00:54:35.869> it's<00:54:36.230> we<00:54:36.800> publish<00:54:37.460> it<00:54:37.640> publicly<00:54:37.670> it 00:54:38.200 --> 00:54:38.210 align:start position:0% fast it's we publish it publicly it 00:54:38.210 --> 00:54:42.220 align:start position:0% fast it's we publish it publicly it should<00:54:38.420> be<00:54:38.510> like<00:54:38.720> 0.2<00:54:39.710> or<00:54:39.740> 0.3<00:54:40.430> seconds<00:54:41.230> 25 00:54:42.220 --> 00:54:42.230 align:start position:0% should be like 0.2 or 0.3 seconds 25 00:54:42.230 --> 00:54:47.920 align:start position:0% should be like 0.2 or 0.3 seconds 25 blocks<00:54:42.950> gets<00:54:43.940> a<00:54:44.119> little<00:54:44.680> a<00:54:45.680> little<00:54:46.339> painful<00:54:46.930> on 00:54:47.920 --> 00:54:47.930 align:start position:0% blocks gets a little a little painful on 00:54:47.930 --> 00:54:50.109 align:start position:0% blocks gets a little a little painful on an<00:54:48.349> FPGA<00:54:48.920> you're<00:54:49.310> still<00:54:49.550> gonna<00:54:49.640> have<00:54:49.880> to<00:54:49.910> go 00:54:50.109 --> 00:54:50.119 align:start position:0% an FPGA you're still gonna have to go 00:54:50.119 --> 00:54:53.230 align:start position:0% an FPGA you're still gonna have to go and<00:54:50.300> we<00:54:50.960> push<00:54:51.170> out<00:54:51.320> the<00:54:51.470> bit<00:54:51.619> stream<00:54:52.240> every 00:54:53.230 --> 00:54:53.240 align:start position:0% and we push out the bit stream every 00:54:53.240 --> 00:54:54.990 align:start position:0% and we push out the bit stream every every<00:54:53.690> few<00:54:53.869> blocks<00:54:54.080> which<00:54:54.349> will<00:54:54.500> be<00:54:54.589> painful 00:54:54.990 --> 00:54:55.000 align:start position:0% every few blocks which will be painful 00:54:55.000 --> 00:54:59.410 align:start position:0% every few blocks which will be painful that<00:54:56.000> will<00:54:56.210> be<00:54:56.359> a<00:54:57.040> three<00:54:58.040> to<00:54:58.070> three<00:54:58.940> to<00:54:59.150> seven 00:54:59.410 --> 00:54:59.420 align:start position:0% that will be a three to three to seven 00:54:59.420 --> 00:55:02.140 align:start position:0% that will be a three to three to seven minute<00:54:59.869> overhead<00:55:00.310> do<00:55:01.310> you<00:55:01.430> know<00:55:01.580> how<00:55:01.730> much<00:55:01.910> CPU 00:55:02.140 --> 00:55:02.150 align:start position:0% minute overhead do you know how much CPU 00:55:02.150 --> 00:55:05.950 align:start position:0% minute overhead do you know how much CPU that<00:55:03.880> computing<00:55:04.880> the<00:55:05.060> kernel<00:55:05.390> code<00:55:05.540> I<00:55:05.630> don't 00:55:05.950 --> 00:55:05.960 align:start position:0% that computing the kernel code I don't 00:55:05.960 --> 00:55:08.050 align:start position:0% that computing the kernel code I don't actually<00:55:06.200> I<00:55:06.619> should<00:55:06.980> I<00:55:07.339> should<00:55:07.369> measure<00:55:07.790> that 00:55:08.050 --> 00:55:08.060 align:start position:0% actually I should I should measure that 00:55:08.060 --> 00:55:10.390 align:start position:0% actually I should I should measure that I<00:55:08.330> know<00:55:08.540> that<00:55:08.750> some<00:55:09.349> of<00:55:09.500> the<00:55:09.619> mining<00:55:09.950> rigs<00:55:10.160> are 00:55:10.390 --> 00:55:10.400 align:start position:0% I know that some of the mining rigs are 00:55:10.400 --> 00:55:13.240 align:start position:0% I know that some of the mining rigs are running<00:55:11.180> on<00:55:11.330> almost<00:55:11.990> full<00:55:12.230> capacity<00:55:12.830> on<00:55:13.040> this 00:55:13.240 --> 00:55:13.250 align:start position:0% running on almost full capacity on this 00:55:13.250 --> 00:55:16.900 align:start position:0% running on almost full capacity on this like<00:55:13.580> really<00:55:14.330> small<00:55:15.609> intercourse 00:55:16.900 --> 00:55:16.910 align:start position:0% like really small intercourse 00:55:16.910 --> 00:55:19.060 align:start position:0% like really small intercourse most<00:55:17.120> of<00:55:17.360> them<00:55:17.480> have<00:55:17.660> a<00:55:17.690> Core<00:55:18.050> 2<00:55:18.290> Duo<00:55:18.320> which 00:55:19.060 --> 00:55:19.070 align:start position:0% most of them have a Core 2 Duo which 00:55:19.070 --> 00:55:21.490 align:start position:0% most of them have a Core 2 Duo which should<00:55:19.370> be<00:55:19.520> okay<00:55:19.880> with<00:55:20.090> it<00:55:20.120> quite<00:55:20.960> frankly<00:55:21.140> and 00:55:21.490 --> 00:55:21.500 align:start position:0% should be okay with it quite frankly and 00:55:21.500 --> 00:55:24.100 align:start position:0% should be okay with it quite frankly and all<00:55:21.920> of<00:55:22.190> our<00:55:22.280> computation<00:55:23.060> as<00:55:23.210> well<00:55:23.450> takes 00:55:24.100 --> 00:55:24.110 align:start position:0% all of our computation as well takes 00:55:24.110 --> 00:55:25.870 align:start position:0% all of our computation as well takes place<00:55:24.410> on<00:55:24.680> the<00:55:24.860> GPU<00:55:25.280> and<00:55:25.460> we<00:55:25.580> can<00:55:25.730> actually 00:55:25.870 --> 00:55:25.880 align:start position:0% place on the GPU and we can actually 00:55:25.880 --> 00:55:29.080 align:start position:0% place on the GPU and we can actually make<00:55:26.270> sure<00:55:26.540> that<00:55:26.810> oh<00:55:27.080> the<00:55:27.890> CPU<00:55:28.520> is<00:55:28.640> just<00:55:28.880> doing 00:55:29.080 --> 00:55:29.090 align:start position:0% make sure that oh the CPU is just doing 00:55:29.090 --> 00:55:31.360 align:start position:0% make sure that oh the CPU is just doing a<00:55:29.240> very<00:55:29.450> slight<00:55:29.720> amount<00:55:29.750> of<00:55:30.050> verification<00:55:30.680> so 00:55:31.360 --> 00:55:31.370 align:start position:0% a very slight amount of verification so 00:55:31.370 --> 00:55:33.070 align:start position:0% a very slight amount of verification so we<00:55:31.490> could<00:55:31.640> actually<00:55:31.760> tune<00:55:32.180> that<00:55:32.330> that's<00:55:32.780> more 00:55:33.070 --> 00:55:33.080 align:start position:0% we could actually tune that that's more 00:55:33.080 --> 00:55:35.140 align:start position:0% we could actually tune that that's more of<00:55:33.200> a<00:55:33.320> that's<00:55:33.920> not<00:55:34.010> on<00:55:34.610> the<00:55:34.760> proof-of-work 00:55:35.140 --> 00:55:35.150 align:start position:0% of a that's not on the proof-of-work 00:55:35.150 --> 00:55:36.490 align:start position:0% of a that's not on the proof-of-work side<00:55:35.540> though<00:55:35.720> that's<00:55:35.960> more<00:55:36.260> in<00:55:36.410> the 00:55:36.490 --> 00:55:36.500 align:start position:0% side though that's more in the 00:55:36.500 --> 00:55:38.950 align:start position:0% side though that's more in the application<00:55:36.650> or<00:55:37.400> like<00:55:37.610> the<00:55:37.760> current<00:55:38.030> the 00:55:38.950 --> 00:55:38.960 align:start position:0% application or like the current the 00:55:38.960 --> 00:55:41.230 align:start position:0% application or like the current the client<00:55:39.350> side<00:55:39.590> so<00:55:40.400> definitely<00:55:40.790> we<00:55:40.940> need<00:55:41.120> to 00:55:41.230 --> 00:55:41.240 align:start position:0% client side so definitely we need to 00:55:41.240 --> 00:55:42.730 align:start position:0% client side so definitely we need to work<00:55:41.390> with<00:55:41.420> the<00:55:41.750> etherium<00:55:42.080> developers<00:55:42.560> for 00:55:42.730 --> 00:55:42.740 align:start position:0% work with the etherium developers for 00:55:42.740 --> 00:55:44.710 align:start position:0% work with the etherium developers for that<00:55:42.860> again<00:55:43.610> the<00:55:43.760> clients<00:55:44.240> are<00:55:44.360> not<00:55:44.540> our 00:55:44.710 --> 00:55:44.720 align:start position:0% that again the clients are not our 00:55:44.720 --> 00:55:48.130 align:start position:0% that again the clients are not our strong<00:55:45.020> suite<00:55:45.320> as<00:55:46.220> as<00:55:46.640> is<00:55:46.940> proven<00:55:47.420> by<00:55:47.570> the<00:55:47.630> cpu 00:55:48.130 --> 00:55:48.140 align:start position:0% strong suite as as is proven by the cpu 00:55:48.140 --> 00:55:53.140 align:start position:0% strong suite as as is proven by the cpu verification<00:55:50.260> will<00:55:51.260> you<00:55:51.380> update<00:55:51.710> the<00:55:51.880> DEP<00:55:52.880> to 00:55:53.140 --> 00:55:53.150 align:start position:0% verification will you update the DEP to 00:55:53.150 --> 00:55:56.050 align:start position:0% verification will you update the DEP to have<00:55:53.750> the<00:55:53.900> new<00:55:54.080> spec<00:55:54.440> yes<00:55:55.070> that<00:55:55.460> has<00:55:55.670> been<00:55:55.880> on 00:55:56.050 --> 00:55:56.060 align:start position:0% have the new spec yes that has been on 00:55:56.060 --> 00:55:58.390 align:start position:0% have the new spec yes that has been on my<00:55:56.210> list<00:55:56.240> of<00:55:56.540> things<00:55:56.600> to<00:55:56.870> do<00:55:57.110> I<00:55:57.320> did<00:55:58.190> get<00:55:58.280> the 00:55:58.390 --> 00:55:58.400 align:start position:0% my list of things to do I did get the 00:55:58.400 --> 00:56:00.700 align:start position:0% my list of things to do I did get the medium<00:55:58.670> article<00:55:59.030> out<00:55:59.180> I<00:55:59.540> did<00:55:59.870> update<00:56:00.230> the<00:56:00.470> I 00:56:00.700 --> 00:56:00.710 align:start position:0% medium article out I did update the I 00:56:00.710 --> 00:56:04.320 align:start position:0% medium article out I did update the I did<00:56:00.950> get<00:56:01.550> hub<00:56:01.760> yes<00:56:02.120> I<00:56:02.300> need<00:56:02.360> to<00:56:02.480> update<00:56:02.720> the<00:56:03.020> EIP 00:56:04.320 --> 00:56:04.330 align:start position:0% did get hub yes I need to update the EIP 00:56:04.330 --> 00:56:07.270 align:start position:0% did get hub yes I need to update the EIP will<00:56:05.330> you<00:56:05.570> develop<00:56:06.320> some<00:56:06.560> test<00:56:06.830> cases<00:56:07.100> for 00:56:07.270 --> 00:56:07.280 align:start position:0% will you develop some test cases for 00:56:07.280 --> 00:56:10.990 align:start position:0% will you develop some test cases for other<00:56:07.880> clients<00:56:08.360> to<00:56:08.630> use<00:56:09.140> yes<00:56:10.010> I<00:56:10.400> need<00:56:10.550> help 00:56:10.990 --> 00:56:11.000 align:start position:0% other clients to use yes I need help 00:56:11.000 --> 00:56:15.040 align:start position:0% other clients to use yes I need help with<00:56:11.180> those<00:56:11.360> quite<00:56:11.960> frankly<00:56:12.290> I<00:56:13.570> think<00:56:14.570> that 00:56:15.040 --> 00:56:15.050 align:start position:0% with those quite frankly I think that 00:56:15.050 --> 00:56:18.490 align:start position:0% with those quite frankly I think that would<00:56:15.260> be<00:56:15.290> amazing<00:56:15.950> I<00:56:16.460> do<00:56:16.850> not<00:56:17.090> have<00:56:17.500> enough 00:56:18.490 --> 00:56:18.500 align:start position:0% would be amazing I do not have enough 00:56:18.500 --> 00:56:21.160 align:start position:0% would be amazing I do not have enough spare<00:56:18.950> cycles<00:56:19.220> to<00:56:20.150> date<00:56:20.360> to<00:56:20.570> be<00:56:20.690> able<00:56:20.810> to<00:56:20.960> do 00:56:21.160 --> 00:56:21.170 align:start position:0% spare cycles to date to be able to do 00:56:21.170 --> 00:56:25.510 align:start position:0% spare cycles to date to be able to do all<00:56:21.560> of<00:56:21.620> these<00:56:21.980> things<00:56:22.270> okay<00:56:23.270> thank<00:56:23.660> you<00:56:24.520> about 00:56:25.510 --> 00:56:25.520 align:start position:0% all of these things okay thank you about 00:56:25.520 --> 00:56:28.380 align:start position:0% all of these things okay thank you about five<00:56:25.700> minutes<00:56:26.000> left 00:56:28.380 --> 00:56:28.390 align:start position:0% 00:56:28.390 --> 00:56:30.570 align:start position:0% could<00:56:29.390> you<00:56:29.510> speak<00:56:29.720> a<00:56:29.840> little<00:56:29.960> more<00:56:30.080> about 00:56:30.570 --> 00:56:30.580 align:start position:0% could you speak a little more about 00:56:30.580 --> 00:56:33.670 align:start position:0% could you speak a little more about whether<00:56:31.580> this<00:56:32.390> changes<00:56:33.170> to<00:56:33.320> bandwidth 00:56:33.670 --> 00:56:33.680 align:start position:0% whether this changes to bandwidth 00:56:33.680 --> 00:56:36.580 align:start position:0% whether this changes to bandwidth consumption<00:56:34.220> on<00:56:34.670> the<00:56:34.700> bus<00:56:35.000> on<00:56:35.270> the<00:56:35.330> other<00:56:36.110> GPU 00:56:36.580 --> 00:56:36.590 align:start position:0% consumption on the bus on the other GPU 00:56:36.590 --> 00:56:40.030 align:start position:0% consumption on the bus on the other GPU because<00:56:37.220> of<00:56:37.400> the<00:56:37.520> increased<00:56:38.030> cache<00:56:38.930> size<00:56:39.230> yeah 00:56:40.030 --> 00:56:40.040 align:start position:0% because of the increased cache size yeah 00:56:40.040 --> 00:56:41.620 align:start position:0% because of the increased cache size yeah that's<00:56:40.610> that's<00:56:40.940> the<00:56:41.120> whole<00:56:41.240> point<00:56:41.450> it 00:56:41.620 --> 00:56:41.630 align:start position:0% that's that's the whole point it 00:56:41.630 --> 00:56:42.880 align:start position:0% that's that's the whole point it actually<00:56:41.930> does<00:56:42.110> change<00:56:42.410> the<00:56:42.470> bandwidth 00:56:42.880 --> 00:56:42.890 align:start position:0% actually does change the bandwidth 00:56:42.890 --> 00:56:46.150 align:start position:0% actually does change the bandwidth consumption<00:56:43.310> again<00:56:43.910> generically<00:56:44.450> it's<00:56:45.160> only 00:56:46.150 --> 00:56:46.160 align:start position:0% consumption again generically it's only 00:56:46.160 --> 00:56:51.100 align:start position:0% consumption again generically it's only 60%<00:56:46.780> consumed<00:56:47.780> which<00:56:48.560> is<00:56:48.710> a<00:56:48.740> problem<00:56:49.510> we<00:56:50.510> do<00:56:50.660> 90 00:56:51.100 --> 00:56:51.110 align:start position:0% 60% consumed which is a problem we do 90 00:56:51.110 --> 00:56:55.630 align:start position:0% 60% consumed which is a problem we do 90 percent<00:56:52.090> you<00:56:53.090> can<00:56:53.330> never<00:56:53.600> get<00:56:53.900> 100%<00:56:54.860> bandwidth 00:56:55.630 --> 00:56:55.640 align:start position:0% percent you can never get 100% bandwidth 00:56:55.640 --> 00:56:57.900 align:start position:0% percent you can never get 100% bandwidth consumption<00:56:56.150> in<00:56:56.420> real<00:56:56.630> world<00:56:56.900> applications 00:56:57.900 --> 00:56:57.910 align:start position:0% consumption in real world applications 00:56:57.910 --> 00:57:00.580 align:start position:0% consumption in real world applications it's<00:56:58.910> just<00:56:59.270> it's<00:56:59.630> impossible<00:57:00.230> it's<00:57:00.350> one<00:57:00.470> of 00:57:00.580 --> 00:57:00.590 align:start position:0% it's just it's impossible it's one of 00:57:00.590 --> 00:57:02.230 align:start position:0% it's just it's impossible it's one of the<00:57:00.650> things<00:57:00.770> GPU<00:57:01.190> architects<00:57:01.700> struggle<00:57:02.120> with 00:57:02.230 --> 00:57:02.240 align:start position:0% the things GPU architects struggle with 00:57:02.240 --> 00:57:05.170 align:start position:0% the things GPU architects struggle with so<00:57:03.020> that's<00:57:03.740> the<00:57:04.070> best<00:57:04.280> we're<00:57:04.520> able<00:57:04.640> to<00:57:04.850> do<00:57:04.970> to 00:57:05.170 --> 00:57:05.180 align:start position:0% so that's the best we're able to do to 00:57:05.180 --> 00:57:09.910 align:start position:0% so that's the best we're able to do to date<00:57:05.440> in<00:57:06.640> theory<00:57:07.640> you<00:57:07.940> could<00:57:08.560> I'm<00:57:09.560> sure 00:57:09.910 --> 00:57:09.920 align:start position:0% date in theory you could I'm sure 00:57:09.920 --> 00:57:13.390 align:start position:0% date in theory you could I'm sure there's<00:57:10.580> like<00:57:10.850> a<00:57:11.030> trick<00:57:11.420> and<00:57:12.170> I'll<00:57:12.350> go<00:57:13.010> noodle 00:57:13.390 --> 00:57:13.400 align:start position:0% there's like a trick and I'll go noodle 00:57:13.400 --> 00:57:15.010 align:start position:0% there's like a trick and I'll go noodle on<00:57:13.550> this<00:57:13.730> because<00:57:13.970> if<00:57:14.210> probe<00:57:14.420> power<00:57:14.660> does<00:57:14.810> get 00:57:15.010 --> 00:57:15.020 align:start position:0% on this because if probe power does get 00:57:15.020 --> 00:57:16.780 align:start position:0% on this because if probe power does get adopted<00:57:15.260> naturally<00:57:16.130> I'll<00:57:16.220> want<00:57:16.460> to<00:57:16.520> go<00:57:16.610> and 00:57:16.780 --> 00:57:16.790 align:start position:0% adopted naturally I'll want to go and 00:57:16.790 --> 00:57:19.240 align:start position:0% adopted naturally I'll want to go and break<00:57:16.910> it<00:57:17.120> again<00:57:17.440> I'm<00:57:18.440> sure<00:57:18.680> there's<00:57:18.920> a<00:57:18.980> 1% 00:57:19.240 --> 00:57:19.250 align:start position:0% break it again I'm sure there's a 1% 00:57:19.250 --> 00:57:21.250 align:start position:0% break it again I'm sure there's a 1% speed-up<00:57:19.940> you<00:57:20.090> could<00:57:20.270> get<00:57:20.450> with<00:57:20.690> some<00:57:20.900> clever 00:57:21.250 --> 00:57:21.260 align:start position:0% speed-up you could get with some clever 00:57:21.260 --> 00:57:24.640 align:start position:0% speed-up you could get with some clever tricks<00:57:21.650> on<00:57:21.830> the<00:57:21.920> minor<00:57:22.130> side<00:57:22.900> but<00:57:23.900> other<00:57:24.410> than 00:57:24.640 --> 00:57:24.650 align:start position:0% tricks on the minor side but other than 00:57:24.650 --> 00:57:26.260 align:start position:0% tricks on the minor side but other than that<00:57:24.710> yes<00:57:25.070> it<00:57:25.370> changes<00:57:25.760> your<00:57:25.910> bandwidth 00:57:26.260 --> 00:57:26.270 align:start position:0% that yes it changes your bandwidth 00:57:26.270 --> 00:57:28.300 align:start position:0% that yes it changes your bandwidth consumption<00:57:26.690> consumes<00:57:27.590> 90%<00:57:28.160> of<00:57:28.220> the 00:57:28.300 --> 00:57:28.310 align:start position:0% consumption consumes 90% of the 00:57:28.310 --> 00:57:28.720 align:start position:0% consumption consumes 90% of the bandwidth 00:57:28.720 --> 00:57:28.730 align:start position:0% bandwidth 00:57:28.730 --> 00:57:30.480 align:start position:0% bandwidth that's<00:57:28.970> basically<00:57:29.600> what<00:57:29.900> we<00:57:30.020> mean<00:57:30.050> by 00:57:30.480 --> 00:57:30.490 align:start position:0% that's basically what we mean by 00:57:30.490 --> 00:57:32.920 align:start position:0% that's basically what we mean by bandwidth<00:57:31.490> utilization<00:57:31.700> that's<00:57:32.600> how<00:57:32.750> much 00:57:32.920 --> 00:57:32.930 align:start position:0% bandwidth utilization that's how much 00:57:32.930 --> 00:57:43.140 align:start position:0% bandwidth utilization that's how much bandwidth<00:57:33.290> is<00:57:33.440> consumed 00:57:43.140 --> 00:57:43.150 align:start position:0% 00:57:43.150 --> 00:57:45.940 align:start position:0% hi<00:57:44.150> I'm<00:57:44.300> still<00:57:44.570> thinking<00:57:44.960> about<00:57:44.990> how<00:57:45.560> to<00:57:45.590> break 00:57:45.940 --> 00:57:45.950 align:start position:0% hi I'm still thinking about how to break 00:57:45.950 --> 00:57:49.570 align:start position:0% hi I'm still thinking about how to break your<00:57:46.160> scheme<00:57:46.480> that's<00:57:47.480> okay<00:57:47.840> man<00:57:48.110> that's<00:57:48.580> okay 00:57:49.570 --> 00:57:49.580 align:start position:0% your scheme that's okay man that's okay 00:57:49.580 --> 00:57:51.760 align:start position:0% your scheme that's okay man that's okay I<00:57:49.610> wish<00:57:49.970> I<00:57:50.270> was<00:57:50.690> deaf<00:57:50.990> and<00:57:51.200> elsewhere<00:57:51.470> here 00:57:51.760 --> 00:57:51.770 align:start position:0% I wish I was deaf and elsewhere here 00:57:51.770 --> 00:57:53.470 align:start position:0% I wish I was deaf and elsewhere here even<00:57:52.070> if<00:57:52.190> they<00:57:52.310> were<00:57:52.370> wearing<00:57:52.580> masks<00:57:53.270> they 00:57:53.470 --> 00:57:53.480 align:start position:0% even if they were wearing masks they 00:57:53.480 --> 00:57:58.990 align:start position:0% even if they were wearing masks they love<00:57:53.750> you<00:57:55.750> so<00:57:56.750> I<00:57:57.110> mean<00:57:57.410> even<00:57:57.560> though<00:57:57.800> you<00:57:58.000> let's 00:57:58.990 --> 00:57:59.000 align:start position:0% love you so I mean even though you let's 00:57:59.000 --> 00:58:00.100 align:start position:0% love you so I mean even though you let's say<00:57:59.150> assume<00:57:59.450> you're<00:57:59.600> using<00:57:59.780> a<00:57:59.960> hundred 00:58:00.100 --> 00:58:00.110 align:start position:0% say assume you're using a hundred 00:58:00.110 --> 00:58:02.860 align:start position:0% say assume you're using a hundred percent<00:58:00.260> of<00:58:00.650> the<00:58:00.830> GPU<00:58:01.370> you<00:58:02.180> still<00:58:02.480> eighty 00:58:02.860 --> 00:58:02.870 align:start position:0% percent of the GPU you still eighty 00:58:02.870 --> 00:58:11.680 align:start position:0% percent of the GPU you still eighty percent<00:58:06.280> you<00:58:07.280> still<00:58:10.120> not<00:58:11.120> using<00:58:11.420> the<00:58:11.510> full 00:58:11.680 --> 00:58:11.690 align:start position:0% percent you still not using the full 00:58:11.690 --> 00:58:13.660 align:start position:0% percent you still not using the full flexibility<00:58:12.200> of<00:58:12.410> the<00:58:12.620> of<00:58:12.890> the<00:58:13.010> GPU<00:58:13.310> so<00:58:13.430> for 00:58:13.660 --> 00:58:13.670 align:start position:0% flexibility of the of the GPU so for 00:58:13.670 --> 00:58:15.880 align:start position:0% flexibility of the of the GPU so for example<00:58:14.120> well<00:58:14.510> why<00:58:14.690> you're<00:58:15.290> not<00:58:15.440> using<00:58:15.680> the 00:58:15.880 --> 00:58:15.890 align:start position:0% example well why you're not using the 00:58:15.890 --> 00:58:17.740 align:start position:0% example well why you're not using the full<00:58:16.100> flexibility<00:58:16.700> of<00:58:16.940> the<00:58:17.060> GPU<00:58:17.330> so<00:58:17.450> for 00:58:17.740 --> 00:58:17.750 align:start position:0% full flexibility of the GPU so for 00:58:17.750 --> 00:58:19.600 align:start position:0% full flexibility of the GPU so for example<00:58:18.140> you<00:58:18.650> know<00:58:18.740> there<00:58:18.890> will<00:58:19.010> be<00:58:19.070> a<00:58:19.160> circuit 00:58:19.600 --> 00:58:19.610 align:start position:0% example you know there will be a circuit 00:58:19.610 --> 00:58:21.310 align:start position:0% example you know there will be a circuit that<00:58:19.700> does<00:58:19.760> addition<00:58:20.480> and<00:58:20.720> then<00:58:20.810> a<00:58:20.870> second<00:58:21.110> as 00:58:21.310 --> 00:58:21.320 align:start position:0% that does addition and then a second as 00:58:21.320 --> 00:58:23.350 align:start position:0% that does addition and then a second as multiplication<00:58:21.710> and<00:58:22.580> then<00:58:22.760> you<00:58:22.910> call<00:58:23.180> them 00:58:23.350 --> 00:58:23.360 align:start position:0% multiplication and then you call them 00:58:23.360 --> 00:58:27.220 align:start position:0% multiplication and then you call them one<00:58:23.600> after<00:58:23.750> the<00:58:23.900> other<00:58:24.550> and<00:58:25.750> the<00:58:26.750> reason<00:58:27.080> why 00:58:27.220 --> 00:58:27.230 align:start position:0% one after the other and the reason why 00:58:27.230 --> 00:58:29.110 align:start position:0% one after the other and the reason why these<00:58:27.290> two<00:58:27.530> circuits<00:58:28.070> would<00:58:28.190> be<00:58:28.370> separated<00:58:28.820> in 00:58:29.110 --> 00:58:29.120 align:start position:0% these two circuits would be separated in 00:58:29.120 --> 00:58:31.300 align:start position:0% these two circuits would be separated in the<00:58:29.240> GPU<00:58:29.720> is<00:58:29.870> so<00:58:30.020> that<00:58:30.200> you<00:58:30.440> can<00:58:30.620> you<00:58:30.950> can<00:58:30.980> mix 00:58:31.300 --> 00:58:31.310 align:start position:0% the GPU is so that you can you can mix 00:58:31.310 --> 00:58:33.220 align:start position:0% the GPU is so that you can you can mix and<00:58:31.430> match<00:58:31.490> but<00:58:32.300> because<00:58:32.660> you're<00:58:32.840> calling 00:58:33.220 --> 00:58:33.230 align:start position:0% and match but because you're calling 00:58:33.230 --> 00:58:36.310 align:start position:0% and match but because you're calling them<00:58:33.410> in<00:58:33.650> a<00:58:33.740> specific<00:58:33.980> order<00:58:34.430> as<00:58:34.870> an<00:58:35.870> ASIC 00:58:36.310 --> 00:58:36.320 align:start position:0% them in a specific order as an ASIC 00:58:36.320 --> 00:58:37.750 align:start position:0% them in a specific order as an ASIC designer<00:58:36.650> what<00:58:37.010> I<00:58:37.040> could<00:58:37.250> do<00:58:37.400> is<00:58:37.580> I<00:58:37.700> could 00:58:37.750 --> 00:58:37.760 align:start position:0% designer what I could do is I could 00:58:37.760 --> 00:58:40.510 align:start position:0% designer what I could do is I could build<00:58:38.030> a<00:58:38.360> clever<00:58:39.080> circuit<00:58:39.560> that<00:58:39.710> does<00:58:39.770> a<00:58:40.130> and<00:58:40.400> B 00:58:40.510 --> 00:58:40.520 align:start position:0% build a clever circuit that does a and B 00:58:40.520 --> 00:58:43.540 align:start position:0% build a clever circuit that does a and B at<00:58:40.700> the<00:58:40.880> same<00:58:41.090> time<00:58:41.420> and<00:58:42.040> uses<00:58:43.040> less<00:58:43.340> energy 00:58:43.540 --> 00:58:43.550 align:start position:0% at the same time and uses less energy 00:58:43.550 --> 00:58:48.610 align:start position:0% at the same time and uses less energy than<00:58:44.380> would<00:58:45.380> be<00:58:45.710> if<00:58:46.130> you<00:58:46.400> first<00:58:47.440> use<00:58:48.440> the<00:58:48.590> a 00:58:48.610 --> 00:58:48.620 align:start position:0% than would be if you first use the a 00:58:48.620 --> 00:58:53.670 align:start position:0% than would be if you first use the a circuit<00:58:49.190> and<00:58:49.370> then<00:58:49.520> the<00:58:49.670> B<00:58:49.790> circuit<00:58:51.400> yes 00:58:53.670 --> 00:58:53.680 align:start position:0% circuit and then the B circuit yes 00:58:53.680 --> 00:58:55.780 align:start position:0% circuit and then the B circuit yes actually<00:58:54.680> that<00:58:54.890> was<00:58:55.220> something<00:58:55.580> I<00:58:55.640> was<00:58:55.760> going 00:58:55.780 --> 00:58:55.790 align:start position:0% actually that was something I was going 00:58:55.790 --> 00:58:58.360 align:start position:0% actually that was something I was going to<00:58:55.940> put<00:58:56.090> in<00:58:56.180> my<00:58:56.330> slides<00:58:56.660> we<00:58:57.020> use<00:58:57.260> a<00:58:57.340> again<00:58:58.340> I 00:58:58.360 --> 00:58:58.370 align:start position:0% to put in my slides we use a again I 00:58:58.370 --> 00:59:00.870 align:start position:0% to put in my slides we use a again I mentioned<00:58:58.940> we<00:58:59.030> use<00:58:59.060> a<00:58:59.240> barrel<00:58:59.690> rotate<00:59:00.380> shift 00:59:00.870 --> 00:59:00.880 align:start position:0% mentioned we use a barrel rotate shift 00:59:00.880 --> 00:59:06.580 align:start position:0% mentioned we use a barrel rotate shift our<00:59:01.900> implementation<00:59:02.900> of<00:59:03.880> rotate<00:59:04.880> is<00:59:05.090> more<00:59:05.590> GPU 00:59:06.580 --> 00:59:06.590 align:start position:0% our implementation of rotate is more GPU 00:59:06.590 --> 00:59:08.980 align:start position:0% our implementation of rotate is more GPU implementations<00:59:07.280> I<00:59:07.400> said<00:59:07.580> I<00:59:07.850> should<00:59:08.060> say<00:59:08.240> it's 00:59:08.980 --> 00:59:08.990 align:start position:0% implementations I said I should say it's 00:59:08.990 --> 00:59:11.800 align:start position:0% implementations I said I should say it's more<00:59:09.230> efficient<00:59:09.890> than<00:59:10.430> any<00:59:10.610> existing<00:59:10.880> ASIC<00:59:11.690> to 00:59:11.800 --> 00:59:11.810 align:start position:0% more efficient than any existing ASIC to 00:59:11.810 --> 00:59:13.570 align:start position:0% more efficient than any existing ASIC to date<00:59:11.990> and<00:59:12.200> any<00:59:12.590> existing<00:59:12.800> fixed<00:59:13.430> function 00:59:13.570 --> 00:59:13.580 align:start position:0% date and any existing fixed function 00:59:13.580 --> 00:59:18.070 align:start position:0% date and any existing fixed function hardware<00:59:14.150> again<00:59:15.050> GPUs<00:59:15.980> are<00:59:16.280> I<00:59:16.940> think<00:59:17.900> most 00:59:18.070 --> 00:59:18.080 align:start position:0% hardware again GPUs are I think most 00:59:18.080 --> 00:59:20.410 align:start position:0% hardware again GPUs are I think most people<00:59:18.380> forget<00:59:18.560> GPUs<00:59:19.250> are<00:59:19.430> really<00:59:19.760> just<00:59:19.940> Asics 00:59:20.410 --> 00:59:20.420 align:start position:0% people forget GPUs are really just Asics 00:59:20.420 --> 00:59:24.340 align:start position:0% people forget GPUs are really just Asics for<00:59:20.660> meth<00:59:20.920> that<00:59:21.920> is<00:59:22.070> what<00:59:22.280> a<00:59:22.310> GPU<00:59:22.700> is<00:59:23.000> so<00:59:23.930> what 00:59:24.340 --> 00:59:24.350 align:start position:0% for meth that is what a GPU is so what 00:59:24.350 --> 00:59:25.980 align:start position:0% for meth that is what a GPU is so what you're<00:59:24.470> specifically<00:59:25.130> referring<00:59:25.460> to<00:59:25.700> is<00:59:25.850> a 00:59:25.980 --> 00:59:25.990 align:start position:0% you're specifically referring to is a 00:59:25.990 --> 00:59:29.590 align:start position:0% you're specifically referring to is a barrel<00:59:26.990> shift<00:59:27.230> rotator<00:59:27.740> and<00:59:27.980> the<00:59:28.940> GPU<00:59:29.300> right 00:59:29.590 --> 00:59:29.600 align:start position:0% barrel shift rotator and the GPU right 00:59:29.600 --> 00:59:31.660 align:start position:0% barrel shift rotator and the GPU right to<00:59:29.930> date<00:59:30.140> is<00:59:30.350> the<00:59:30.530> most<00:59:30.710> efficient<00:59:31.190> efficient 00:59:31.660 --> 00:59:31.670 align:start position:0% to date is the most efficient efficient 00:59:31.670 --> 00:59:33.820 align:start position:0% to date is the most efficient efficient implementation<00:59:32.330> of<00:59:32.390> it<00:59:32.570> that's<00:59:33.110> why<00:59:33.290> when<00:59:33.590> I 00:59:33.820 --> 00:59:33.830 align:start position:0% implementation of it that's why when I 00:59:33.830 --> 00:59:35.710 align:start position:0% implementation of it that's why when I say<00:59:34.100> that<00:59:34.430> you<00:59:34.910> know<00:59:35.030> if<00:59:35.150> we<00:59:35.270> want<00:59:35.450> to<00:59:35.510> go<00:59:35.600> and 00:59:35.710 --> 00:59:35.720 align:start position:0% say that you know if we want to go and 00:59:35.720 --> 00:59:37.900 align:start position:0% say that you know if we want to go and make<00:59:35.870> a<00:59:35.930> probe<00:59:36.560> by<00:59:36.740> a<00:59:36.920> sikh<00:59:37.160> it<00:59:37.310> just<00:59:37.580> ends<00:59:37.850> up 00:59:37.900 --> 00:59:37.910 align:start position:0% make a probe by a sikh it just ends up 00:59:37.910 --> 00:59:40.030 align:start position:0% make a probe by a sikh it just ends up mimicking<00:59:38.360> making<00:59:39.050> a<00:59:39.140> GPU<00:59:39.470> because<00:59:39.740> you<00:59:39.830> just 00:59:40.030 --> 00:59:40.040 align:start position:0% mimicking making a GPU because you just 00:59:40.040 --> 00:59:42.400 align:start position:0% mimicking making a GPU because you just mimic<00:59:40.310> that<00:59:40.460> i'd<00:59:41.270> love<00:59:41.510> to<00:59:41.540> touch<00:59:41.780> more 00:59:42.400 --> 00:59:42.410 align:start position:0% mimic that i'd love to touch more 00:59:42.410 --> 00:59:44.170 align:start position:0% mimic that i'd love to touch more you<00:59:42.530> after<00:59:42.859> this<00:59:43.010> on<00:59:43.190> that<00:59:43.309> actually<00:59:43.700> because 00:59:44.170 --> 00:59:44.180 align:start position:0% you after this on that actually because 00:59:44.180 --> 00:59:46.059 align:start position:0% you after this on that actually because I<00:59:44.210> have<00:59:44.420> a<00:59:44.450> write-up<00:59:44.869> on<00:59:45.020> that<00:59:45.230> because<00:59:45.740> we<00:59:45.829> got 00:59:46.059 --> 00:59:46.069 align:start position:0% I have a write-up on that because we got 00:59:46.069 --> 00:59:48.309 align:start position:0% I have a write-up on that because we got that<00:59:46.099> question<00:59:46.640> asked<00:59:46.940> a<00:59:46.970> lot<00:59:47.180> from<00:59:47.510> an<00:59:47.780> FPGA 00:59:48.309 --> 00:59:48.319 align:start position:0% that question asked a lot from an FPGA 00:59:48.319 --> 00:59:51.880 align:start position:0% that question asked a lot from an FPGA developer<00:59:49.839> who<00:59:50.839> was<00:59:51.049> who<00:59:51.319> was<00:59:51.349> talking<00:59:51.799> about 00:59:51.880 --> 00:59:51.890 align:start position:0% developer who was who was talking about 00:59:51.890 --> 00:59:53.710 align:start position:0% developer who was who was talking about putting<00:59:52.309> in<00:59:52.579> square<00:59:52.910> roots<00:59:53.119> and<00:59:53.299> other<00:59:53.420> things 00:59:53.710 --> 00:59:53.720 align:start position:0% putting in square roots and other things 00:59:53.720 --> 01:00:03.549 align:start position:0% putting in square roots and other things into<00:59:53.990> propel<00:59:54.760> so<01:00:01.059> I'm<01:00:02.059> done<01:00:02.359> you're<01:00:02.960> releasing 01:00:03.549 --> 01:00:03.559 align:start position:0% into propel so I'm done you're releasing 01:00:03.559 --> 01:00:15.340 align:start position:0% into propel so I'm done you're releasing me<01:00:03.829> from<01:00:03.859> my<01:00:04.130> prison 01:00:15.340 --> 01:00:15.350 align:start position:0% 01:00:15.350 --> 01:00:17.410 align:start position:0% you