1 00:00:00,540 --> 00:00:01,300 All right. 2 00:00:01,300 --> 00:00:04,590 Now we're going to talk about what hardware miners use to achieve this 3 00:00:04,590 --> 00:00:07,500 computation, which I've just spent time telling you is really, 4 00:00:07,500 --> 00:00:08,500 really difficult to do. 5 00:00:10,220 --> 00:00:13,010 So, first of all, what exactly is this computation? 6 00:00:14,170 --> 00:00:18,053 So, we've mentioned hash functions and we've mentioned SHA-256. 7 00:00:18,053 --> 00:00:22,140 In particular, what exactly is SHA-256? 8 00:00:22,140 --> 00:00:24,650 Well, it's a general purpose cryptographic hash function. 9 00:00:24,650 --> 00:00:29,340 It's actually part of a bigger family of functions, that was standardized in 2001. 10 00:00:29,340 --> 00:00:34,240 It did come out of the NSA, which has led to a couple of interesting 11 00:00:34,240 --> 00:00:38,560 conspiracy theories, by some people about its relationship to Bitcoin. 12 00:00:38,560 --> 00:00:42,040 But it's generally considered a fairly strong hash function. 13 00:00:42,040 --> 00:00:43,480 It's not broken, cryptographically. 14 00:00:43,480 --> 00:00:47,220 Although there are some theoretical weaknesses that are starting to show up. 15 00:00:47,220 --> 00:00:52,750 As a result, the replacement the SHA-3 family, has already been picked, actually. 16 00:00:52,750 --> 00:00:55,750 And it's in the final stages of standardization today. 17 00:00:55,750 --> 00:00:58,650 But it wasn't available at the time, BitCoin was designed. 18 00:00:58,650 --> 00:00:59,620 This was a good choice. 19 00:00:59,620 --> 00:01:02,300 This was the strongest general purpose cryptographic hash 20 00:01:02,300 --> 00:01:04,380 function available at the time. 21 00:01:04,380 --> 00:01:08,640 It is possible that it will become less secure over the lifetime of BitCoin, but 22 00:01:08,640 --> 00:01:09,590 for now it's pretty good. 23 00:01:11,500 --> 00:01:12,400 So, what did it look like? 24 00:01:12,400 --> 00:01:17,250 This is a diagram of what the SHA-256 state looks like. 25 00:01:18,660 --> 00:01:21,420 And we don't need to know all of the details of this to understand how 26 00:01:21,420 --> 00:01:22,770 BitCoin works. 27 00:01:22,770 --> 00:01:25,098 But I'll just give a high level overview, 28 00:01:25,098 --> 00:01:28,533 to give an idea of the task that needs to be solved by the miners. 29 00:01:30,805 --> 00:01:34,210 So there's a 256-bit-state in SHA-256. 30 00:01:34,210 --> 00:01:37,700 It's split up into eight 32 bit words. 31 00:01:37,700 --> 00:01:40,809 So this is very much optimized for 32 bit platforms. 32 00:01:43,500 --> 00:01:46,030 And in each round some of those words are taken. 33 00:01:46,030 --> 00:01:49,440 There's four different tweaks that are applied to those words that are at the bit 34 00:01:49,440 --> 00:01:51,240 level, so each one of these is two or 35 00:01:51,240 --> 00:01:56,150 three bits being flipped or ended together, basic logic operations. 36 00:01:57,910 --> 00:02:00,790 And then a number of words in the state are taken, 37 00:02:00,790 --> 00:02:04,454 some with these tweaks applied, and added together mod 32. 38 00:02:05,520 --> 00:02:06,430 In a pipeline here. 39 00:02:06,430 --> 00:02:11,450 And then the result of all of these additions [COUGH] 40 00:02:11,450 --> 00:02:17,750 is wired over to the first word of the state, and the entire state shifts over. 41 00:02:17,750 --> 00:02:20,500 So some of the design here, some of the design ideas, 42 00:02:20,500 --> 00:02:24,070 date back to classic linear feedback shift registers. 43 00:02:24,070 --> 00:02:27,050 Which was one of the earliest approaches to cryptographic design. 44 00:02:29,160 --> 00:02:33,240 So all of this happens, this is just one round of the shock compression function, 45 00:02:34,400 --> 00:02:38,790 and the complete computation of SHA-256 does this 80 times, for 46 00:02:38,790 --> 00:02:43,830 80 iterations, and in each iteration, there are slightly different constants 47 00:02:43,830 --> 00:02:47,990 applied at the one step there, so that every iteration isn't exactly the same. 48 00:02:50,200 --> 00:02:51,730 So, this is the task ahead of miners, 49 00:02:51,730 --> 00:02:56,200 to compute this specific function, only this, as fast as possible. 50 00:02:57,340 --> 00:03:00,970 And so, you can see that to do this you need to be able to deal with 32 bit words, 51 00:03:00,970 --> 00:03:03,940 you need to be able to do 32 bit addition. 52 00:03:03,940 --> 00:03:06,674 And you need to also be able to do some bitwise logic. 53 00:03:09,088 --> 00:03:13,043 So the first generation of mining, when Bitcoin was originally proposed, 54 00:03:13,043 --> 00:03:15,440 was all done on general purpose computers. 55 00:03:15,440 --> 00:03:17,920 General purpose CPUs. 56 00:03:17,920 --> 00:03:22,010 In fact, it was as simple as this code, here. 57 00:03:22,010 --> 00:03:26,340 That simply searched over nonces in a linear fashion, 58 00:03:26,340 --> 00:03:30,180 computed SHA256 in software, and checked if the result was a valid block. 59 00:03:32,510 --> 00:03:36,570 And one quirk which I haven't mentioned yet, this is probably a good time 60 00:03:36,570 --> 00:03:41,150 to mention it if there ever was one, is that for reasons that aren't 61 00:03:41,150 --> 00:03:45,080 completely specified, you actually compute the SHA256 function twice. 62 00:03:46,970 --> 00:03:50,346 So it's just doubled up, that's a fact of life that miners have to live with. 63 00:03:52,725 --> 00:03:56,549 So how fast will this run on a general purpose computer? 64 00:03:57,790 --> 00:03:59,280 Well, if you're doing pretty well. 65 00:03:59,280 --> 00:04:01,240 If you have a high end desktop PC, 66 00:04:01,240 --> 00:04:06,780 you might be able to do this at about two to the 24 hashes per second. 67 00:04:08,680 --> 00:04:12,230 Which would be about 20 megahertz. 68 00:04:13,940 --> 00:04:17,360 Remember that even though hertz often are applied for processor speed, 69 00:04:17,360 --> 00:04:21,240 the basic idea of hertz is it just means something that you're doing, 70 00:04:21,240 --> 00:04:22,600 that many times per second. 71 00:04:24,290 --> 00:04:27,010 Okay, so if you're hashing at a rate of 20 megahertz, 72 00:04:27,010 --> 00:04:28,950 what does that get you with today's difficulty? 73 00:04:32,840 --> 00:04:35,790 Well, it would actually take you over 100,000 74 00:04:35,790 --> 00:04:37,389 years at this rate to find a block. 75 00:04:39,450 --> 00:04:43,110 So we talked about how mining was going to be a difficult slog. 76 00:04:43,110 --> 00:04:47,060 If you're mining on a general purpose PC today, it's a really, 77 00:04:47,060 --> 00:04:48,190 really big hill to get up. 78 00:04:49,370 --> 00:04:52,030 because it's going to take you that 140,000 years. 79 00:04:53,310 --> 00:04:57,416 So as a result, today and even for the last few years, anybody trying to do 80 00:04:57,416 --> 00:05:01,392 mining on a CPU, probably didn't understand how Bitcoin worked, and 81 00:05:01,392 --> 00:05:05,654 were probably pretty disappointed that they never made any money doing it. 82 00:05:07,962 --> 00:05:11,900 So CPU mining was the first generation of mining. 83 00:05:11,900 --> 00:05:13,170 The second generation, 84 00:05:13,170 --> 00:05:17,091 when people started to get frustrated with how slow their CPUs were. 85 00:05:17,091 --> 00:05:21,780 Let's just try to use the graphics card, the graphics processing unit, or GPU. 86 00:05:23,850 --> 00:05:25,550 So what is a GPU? 87 00:05:25,550 --> 00:05:30,540 Almost every computer now has a GPU built in, for high performance graphics. 88 00:05:30,540 --> 00:05:34,846 They're designed to have high parallelism which does help with bitcoin mining, 89 00:05:34,846 --> 00:05:38,680 because you can parallelize and computer multiple hashes at the same time for 90 00:05:38,680 --> 00:05:40,240 different nonces that you want to try. 91 00:05:40,240 --> 00:05:42,980 And they're also designed for high throughput, so 92 00:05:42,980 --> 00:05:45,850 there's a highly pipelined design in graphics cards. 93 00:05:47,350 --> 00:05:48,550 And around 2010, 94 00:05:48,550 --> 00:05:52,670 the first implementation came out, written in a language called OpenCL, 95 00:05:52,670 --> 00:05:57,220 which is a general purpose language to do things other than graphics on a GPU. 96 00:05:58,230 --> 00:06:01,840 and it's a high level language, so it took a little while before people started 97 00:06:01,840 --> 00:06:05,480 tweaking the code even further to run quickly on specific graphics cards. 98 00:06:07,910 --> 00:06:10,240 So what's the advantage of using a graphics card? 99 00:06:10,240 --> 00:06:13,500 Well, [COUGH] for one thing, they're easily available and they're easy for 100 00:06:13,500 --> 00:06:14,850 amateurs to set up. 101 00:06:14,850 --> 00:06:17,420 You can order graphics cards online. 102 00:06:17,420 --> 00:06:21,990 You can buy them now at most big consumer electronic stores. 103 00:06:21,990 --> 00:06:26,374 So they're just the most accessible high end hardware that's available to most 104 00:06:26,374 --> 00:06:26,894 people. 105 00:06:28,299 --> 00:06:32,060 They also have some properties that make them specifically nice for Bitcoin. 106 00:06:32,060 --> 00:06:37,700 They're designed for parallelism so they have a lot of arithmetic logic units 107 00:06:37,700 --> 00:06:42,800 they all use that you can use in parallel to do different shaw 256 computations. 108 00:06:42,800 --> 00:06:46,970 And some of them also have some specific instructions to do bit fiddling 109 00:06:46,970 --> 00:06:48,720 that work out quite nicely for shaw 256. 110 00:06:48,720 --> 00:06:53,550 They also have the property that you can drive many graphics cards from 111 00:06:53,550 --> 00:06:57,760 one motherboard in the CPU, so you could take your one computer and 112 00:06:57,760 --> 00:07:00,350 attach multiple graphics cards to it, if you want. 113 00:07:01,670 --> 00:07:03,880 Most graphics cards can also be overclocked, 114 00:07:03,880 --> 00:07:05,700 which is a property that gamers demand. 115 00:07:05,700 --> 00:07:09,740 So you can run them faster than they're actually designed for, 116 00:07:09,740 --> 00:07:11,140 if you want to take the risk on. 117 00:07:12,940 --> 00:07:17,210 And with bitcoin mining, it might be a good idea to run the chip must faster than 118 00:07:17,210 --> 00:07:20,910 it was designed for, even if you introduce some errors into the process. 119 00:07:22,200 --> 00:07:27,550 So there's a measure of mining success called Goodput, which is the throughput, 120 00:07:27,550 --> 00:07:29,910 how quickly you're finding blocks. 121 00:07:29,910 --> 00:07:31,165 Time's the success rate. 122 00:07:31,165 --> 00:07:34,320 [COUGH] How often does the computation actually have errors? 123 00:07:35,910 --> 00:07:40,637 An interesting observation, is that it may be worthwhile to go much faster, 124 00:07:40,637 --> 00:07:43,182 even if you make a large number of errors, 125 00:07:43,182 --> 00:07:46,260 if it leads you to find valid blocks more quickly. 126 00:07:46,260 --> 00:07:51,254 So [COUGH] if you can tweak one knob and run your graphics card 50% faster, 127 00:07:51,254 --> 00:07:56,327 even if you have an error in the SHA-256 computation 30% of the time, 128 00:07:56,327 --> 00:08:01,085 you just multiply 1.5 times 0.7 and you're still finding blocks 129 00:08:01,085 --> 00:08:04,940 faster than running the chip at normal speed with error. 130 00:08:04,940 --> 00:08:08,060 So this is something that people spent a long time on optimizing, 131 00:08:08,060 --> 00:08:12,730 exactly how much did they overclock the chip, and what errors did it introduce. 132 00:08:16,090 --> 00:08:18,218 So people started trying to scale this up. 133 00:08:18,218 --> 00:08:19,890 They said, aha! 134 00:08:19,890 --> 00:08:23,390 I have a graphics card and I'm mining faster than I was in my CPU. 135 00:08:23,390 --> 00:08:25,270 What's better than one graphics card? 136 00:08:25,270 --> 00:08:27,330 Lots of graphics cards. 137 00:08:27,330 --> 00:08:30,927 And you started to see these really interesting home grew setups like this one 138 00:08:30,927 --> 00:08:34,110 here, where people hand built their own racks. 139 00:08:34,110 --> 00:08:36,480 They had their own custom cooling setup, 140 00:08:36,480 --> 00:08:41,150 which is often as simple as buying a bunch of fans or air conditioning units. 141 00:08:41,150 --> 00:08:44,820 And trying to run as many GPUs as possible in their basement or 142 00:08:44,820 --> 00:08:46,850 whatever other room was available to them. 143 00:08:48,320 --> 00:08:52,667 So this was obviously in the early days of bitcoin when it was still mostly hobbyist 144 00:08:52,667 --> 00:08:56,825 who didn't know a lot about running a modern data center, but they came up with 145 00:08:56,825 --> 00:09:01,109 some quite ingenious designs to pack A lot of graphics cards into a small place and 146 00:09:01,109 --> 00:09:02,322 try to keep them cool. 147 00:09:05,401 --> 00:09:08,680 Now what are the disadvantages of doing this? 148 00:09:08,680 --> 00:09:13,430 Well, GPUs actually have a lot of other hardware specifically for doing video. 149 00:09:14,430 --> 00:09:17,160 Specifically, they have floating point units that you 150 00:09:17,160 --> 00:09:19,370 don't use at all in SHA-256. 151 00:09:19,370 --> 00:09:23,620 So you're actually wasting a lot of hardware from what the factory built. 152 00:09:25,850 --> 00:09:28,990 They also don't have the greatest cooling characteristics when you 153 00:09:28,990 --> 00:09:31,160 put a lot of graphics cards next to one another. 154 00:09:31,160 --> 00:09:35,080 They're not designed to be run all in a row like I showed on a previous slide. 155 00:09:35,080 --> 00:09:37,430 They're designed to be run one graphics card and 156 00:09:37,430 --> 00:09:39,630 one box doing graphics for one computer. 157 00:09:41,530 --> 00:09:43,980 They can also have a fairly large power draw, so 158 00:09:43,980 --> 00:09:47,350 a lot of electricity is being used by these relative to a computer. 159 00:09:49,120 --> 00:09:52,330 And initially there was the problem that you had to build your own board or 160 00:09:52,330 --> 00:09:55,700 buy expensive boards to actually house multiple graphics cards. 161 00:09:57,280 --> 00:09:59,770 So what's the upshot of this, what kind of performance can you get? 162 00:10:01,590 --> 00:10:06,327 Well, on a good card, a really high-end graphics card with a lot of aggressive 163 00:10:06,327 --> 00:10:09,029 tuning, you might get as high as 200 MHz. 164 00:10:11,572 --> 00:10:15,050 Which is about 2 to the 27 hashes per second. 165 00:10:16,370 --> 00:10:19,710 So that's up to an order of magnitude better than you would be doing in 166 00:10:19,710 --> 00:10:20,599 the CPU case. 167 00:10:22,220 --> 00:10:24,870 But even with that improved performance and 168 00:10:24,870 --> 00:10:26,630 even if you're really aggressive and you say. 169 00:10:26,630 --> 00:10:31,250 I want to get 100 graphics cards together for my mining rig, that setup 170 00:10:31,250 --> 00:10:35,220 would still take you 174 years to find the block at the current difficulty. 171 00:10:37,310 --> 00:10:41,020 So, it was a fun era while it lasted, but graphics cards for 172 00:10:41,020 --> 00:10:43,170 BitCoin mining is basically dead. 173 00:10:45,760 --> 00:10:47,410 So, what replaced graphics cards? 174 00:10:48,530 --> 00:10:54,780 Well, around 2011, people started to use FPGAs, or Field Programmable Gate Arrays. 175 00:10:56,560 --> 00:11:00,200 That was around the time that the first implementation of BitCoin mining came out 176 00:11:00,200 --> 00:11:01,120 in the Verilog, 177 00:11:01,120 --> 00:11:04,010 which is the hardware design language that's used to program FPGAs. 178 00:11:06,020 --> 00:11:09,680 And FPGA's are maybe something that you have programmed before if you've taken 179 00:11:09,680 --> 00:11:11,950 a hardware design course. 180 00:11:11,950 --> 00:11:17,480 They are designed for, to have hardware like performance but to have 181 00:11:17,480 --> 00:11:21,830 customers or the owner of the card being able to customize it or reconfigure it in 182 00:11:21,830 --> 00:11:25,970 the field, unlike a chip which is made in a factory and does the same thing forever. 183 00:11:27,910 --> 00:11:30,500 So they do offer better performance than graphics cards, 184 00:11:30,500 --> 00:11:33,140 particularly on some of the Bit fiddling operations. 185 00:11:33,140 --> 00:11:36,160 That kind of stuff is very easy to spec out an FPGA. 186 00:11:37,820 --> 00:11:39,390 And if you know what you're doing, 187 00:11:39,390 --> 00:11:42,490 you can get the cooling to work out better with an FPGA. 188 00:11:42,490 --> 00:11:45,380 You're wasting a little bit less of the card than you would be with 189 00:11:45,380 --> 00:11:46,990 a graphics card. 190 00:11:46,990 --> 00:11:50,960 And you can pack more of these together and drive them from one central unit. 191 00:11:52,180 --> 00:11:53,860 So just like with the graphics cards, 192 00:11:53,860 --> 00:11:58,600 people said once I have the FPGA working why don't I try getting a lot of FPGA's. 193 00:11:58,600 --> 00:12:04,960 I know this doesn't look quite as messy as the graphics card setup looked. 194 00:12:04,960 --> 00:12:09,670 This is a relatively neat rack with neat wires You don't see the cooling set up 195 00:12:09,670 --> 00:12:16,830 here, you still needed a cooling set up with air conditioning or with fans. 196 00:12:16,830 --> 00:12:20,890 But it was possible to build a big array of FPGAs a little bit more neatly and 197 00:12:20,890 --> 00:12:23,394 cleanly then you could with graphics cards. 198 00:12:25,411 --> 00:12:30,312 The disadvantages of using FPGAs are that they were being driven harder for 199 00:12:30,312 --> 00:12:34,822 BitCoin mining by being on all the time and working as hard as possible 200 00:12:34,822 --> 00:12:39,500 than a lot consumer-grade FPGAs were really designed for. 201 00:12:39,500 --> 00:12:41,580 So a lot of people found a number of errors and 202 00:12:41,580 --> 00:12:45,370 malfunctions in their FPGAs as they were doing Bitcoin mining. 203 00:12:45,370 --> 00:12:51,360 It was also, it turned out, to be difficult to optimize the 32-bit ad step, 204 00:12:51,360 --> 00:12:54,000 which as we said is critical for doing SHA256. 205 00:12:54,000 --> 00:12:58,810 They were also just less accessible for people. 206 00:12:58,810 --> 00:13:02,660 It's harder to buy an FPGA, you can't buy one at most stores. 207 00:13:02,660 --> 00:13:05,060 There are fewer people who know how to program FPGA's, 208 00:13:05,060 --> 00:13:06,089 who know how to set them up. 209 00:13:07,310 --> 00:13:11,870 And it turned out that the costs, even though the performance on FPGA's went up, 210 00:13:11,870 --> 00:13:14,490 the cost per performance was only an incremental, 211 00:13:14,490 --> 00:13:17,690 very marginal gain over using graphics cards. 212 00:13:17,690 --> 00:13:20,040 So it was a pretty short lived reign. 213 00:13:20,040 --> 00:13:22,500 Whereas people were mining on graphics cards and 214 00:13:22,500 --> 00:13:26,370 graphics cards were king for maybe a year or so. 215 00:13:26,370 --> 00:13:30,320 It was a much shorter matter of months where FGPAs were a popular solution for 216 00:13:30,320 --> 00:13:31,860 BitCoin mining. 217 00:13:31,860 --> 00:13:33,720 But, if you were using an FPGA and 218 00:13:33,720 --> 00:13:36,660 using it well, you might get up to about a gigahertz. 219 00:13:37,880 --> 00:13:39,950 And now it sounds like we're making some real progress. 220 00:13:39,950 --> 00:13:41,669 We're doing a billion hashes per second. 221 00:13:43,610 --> 00:13:49,110 But even with a one gigahertz setup on a FPGA, and even if you 100 boards. 222 00:13:49,110 --> 00:13:51,820 So again even, if you bought a lot of FPGAs and 223 00:13:51,820 --> 00:13:54,520 shove them all into your amateur mining rig, 224 00:13:54,520 --> 00:13:59,850 it would still take you 25 years to find a BitCoin block at that rate. 225 00:13:59,850 --> 00:14:03,180 So this is still not looking like a really attractive thing to jump into. 226 00:14:04,210 --> 00:14:07,350 And you might be asking if all these solutions are so intractable today, 227 00:14:07,350 --> 00:14:08,780 what are people actually doing? 228 00:14:11,010 --> 00:14:15,030 Well mining today is essentially dominated by BitCoin ASICs, or 229 00:14:15,030 --> 00:14:17,280 Application Specific Integrated Circuits. 230 00:14:17,280 --> 00:14:19,220 So these are chips that were designed and 231 00:14:19,220 --> 00:14:22,719 built from scratch to do nothing except mine BitCoins. 232 00:14:23,980 --> 00:14:29,120 They were fabbed out at a factory, packaged up and sold to a consumer, 233 00:14:29,120 --> 00:14:32,750 solely to be BitCoin miners, and if you go online today, 234 00:14:32,750 --> 00:14:37,310 you can find a lot of people willing to sell you BitCoin mining ASICs. 235 00:14:37,310 --> 00:14:41,010 There's a couple of big vendors that now do this as a full time product, so 236 00:14:41,010 --> 00:14:44,040 these companies have sprung up in the last two years. 237 00:14:44,040 --> 00:14:48,360 And the main product, in some cases the only product that they sell as a business 238 00:14:48,360 --> 00:14:50,010 is Bitcoin-mining ASICs. 239 00:14:52,240 --> 00:14:55,120 And now you'll notice when you're buying an ASIC, you have a lot of options. 240 00:14:55,120 --> 00:14:58,130 You can choose between slightly bigger and more expensive models, 241 00:14:58,130 --> 00:15:00,420 more compact models. 242 00:15:00,420 --> 00:15:03,570 And they'll throw a lot of numbers at you in terms of what the performance is going 243 00:15:03,570 --> 00:15:08,570 to be, the cost, how much power it's going to use But the most important thing 244 00:15:08,570 --> 00:15:12,420 to look at in a lot of these cases is how quickly they're going to ship to you. 245 00:15:14,170 --> 00:15:17,000 So a lot of ASICs have a pretty strong disclaimer 246 00:15:17,000 --> 00:15:19,900 that you have to pre-order them before they're even available and 247 00:15:19,900 --> 00:15:23,140 they don't make any firm guarantees on when they're going to be delivered. 248 00:15:24,170 --> 00:15:27,500 So again since these are new companies selling the ASICs. 249 00:15:27,500 --> 00:15:31,660 It had to be funded as they went, which means they need to essentially pay for 250 00:15:31,660 --> 00:15:36,230 the production run of ASICs with consumers' pre-orders. 251 00:15:36,230 --> 00:15:38,700 And that means there's a lot of pressure on people, 252 00:15:38,700 --> 00:15:41,620 on consumers to pay before the chip is ready. 253 00:15:41,620 --> 00:15:44,404 In a lot of cases, the chips had been shipped late and 254 00:15:44,404 --> 00:15:48,470 this has caused considerable consternation and heartache for customers. 255 00:15:48,470 --> 00:15:51,716 There are a lot of dramatic complaints and stories and 256 00:15:51,716 --> 00:15:55,467 tales of woe of people who spent good money on a BitCoin ASIC and 257 00:15:55,467 --> 00:15:59,600 were hoping to really strike it rich when the thing came in the mail. 258 00:15:59,600 --> 00:16:02,220 Only it came in the mail later than they were expected. 259 00:16:05,020 --> 00:16:06,620 Like I said, these are special purpose. 260 00:16:06,620 --> 00:16:09,980 They're designed to be run at full speed, constantly, for life, 261 00:16:09,980 --> 00:16:12,950 to do nothing other than mine BitCoins. 262 00:16:12,950 --> 00:16:16,470 They require a lot of expertise and a long lead time to design. 263 00:16:16,470 --> 00:16:19,950 So much more complicated to actually tape out a chip than to just write 264 00:16:19,950 --> 00:16:23,890 an implementation in open CLor in Verilog. 265 00:16:23,890 --> 00:16:28,940 But the amazing thing about BitCoin ASICs is that, as hard as they were to design, 266 00:16:28,940 --> 00:16:33,700 analysts have looked at this and said this may be the fastest turn around time 267 00:16:33,700 --> 00:16:38,250 essentially in the history of integrated circuits for specifying a problem, which 268 00:16:38,250 --> 00:16:43,420 was mining BitCoins, and turning it around to have a working chip in people's hands. 269 00:16:44,630 --> 00:16:46,120 So, this was really a rush job. 270 00:16:46,120 --> 00:16:49,300 People realized the need to have BitCoin mining ASICs and 271 00:16:49,300 --> 00:16:51,630 the fact that they could sell them for a lot of money. 272 00:16:51,630 --> 00:16:56,530 They designed them extremely quickly and started shipping them to consumers. 273 00:16:56,530 --> 00:16:59,690 And as a result, you can probably expect there were a lot of bugs in the first few 274 00:16:59,690 --> 00:17:01,430 generations of these things. 275 00:17:01,430 --> 00:17:03,120 Some of them didn't deliver. 276 00:17:03,120 --> 00:17:06,490 In fact, most of them probably didn't deliver quite the performance that they 277 00:17:06,490 --> 00:17:07,880 were promising. 278 00:17:07,880 --> 00:17:10,230 Some of them, in fact, are quite buggy. 279 00:17:10,230 --> 00:17:13,424 But it's evolved a lot over the past year and 280 00:17:13,424 --> 00:17:17,235 they're now fairly reliable ASICs being shipped. 281 00:17:17,235 --> 00:17:20,079 So what does it look like if you actually want to buy one of these things? 282 00:17:21,090 --> 00:17:23,620 So here's just a case study of something that's been available for 283 00:17:23,620 --> 00:17:27,220 about six months now, so it's already a little bit out of date, but 284 00:17:27,220 --> 00:17:31,550 this TerraMiner IV is this nice, big, fancy box you see here. 285 00:17:31,550 --> 00:17:35,960 And this hashes at about two terahertz, so 286 00:17:35,960 --> 00:17:39,410 it's 1,000 times faster than that hypothetical array of 100 really good 287 00:17:39,410 --> 00:17:42,059 FPGAs that we were talking about previously. 288 00:17:43,190 --> 00:17:46,730 So you have to pay about $6,000 for this. 289 00:17:46,730 --> 00:17:51,330 And even with this incredible performance, it would still take 14 months, 290 00:17:51,330 --> 00:17:52,630 on average, to find a block. 291 00:17:53,840 --> 00:17:57,820 So even if you think that you're investing in a fairly nice piece of hardware, 292 00:17:57,820 --> 00:18:00,368 it's still extremely difficult to mine a block. 293 00:18:02,851 --> 00:18:05,380 And the market dynamics are really interesting here. 294 00:18:06,560 --> 00:18:09,950 So most boards that have been coming on the market since 295 00:18:09,950 --> 00:18:14,550 ASICs came out have been effectively obsolete in maybe six months. 296 00:18:15,750 --> 00:18:20,830 And on top of being obsolete in six months the bulk of the profits are made up front, 297 00:18:20,830 --> 00:18:23,960 often in the first six weeks is when you'll make half of expected 298 00:18:23,960 --> 00:18:26,430 profits in the lifetime of a mining rig. 299 00:18:26,430 --> 00:18:30,780 The fact that such a high proportion of the profits are made in the first six 300 00:18:30,780 --> 00:18:34,570 months means that there's an incredible premium on shipping speed. 301 00:18:34,570 --> 00:18:41,370 If your thing ships a week late, you may have lost one sixth of that optimum time. 302 00:18:41,370 --> 00:18:46,060 In fact you have lost the most valuable week in the lifetime of the ASIC because 303 00:18:46,060 --> 00:18:50,210 its value's only going to down over time as mining gets more and more difficult. 304 00:18:51,720 --> 00:18:56,353 And this is really the motivation for why so many companies require preorders, 305 00:18:56,353 --> 00:19:00,240 the fact that there's so much competition to get the ASICs first. 306 00:19:00,240 --> 00:19:04,180 And there's a lot of speculation that some companies have actually manufactured 307 00:19:04,180 --> 00:19:06,290 the ASICs, tried to run them themselves for 308 00:19:06,290 --> 00:19:10,020 a couple of weeks before shipping them, and then ship them off to consumers. 309 00:19:11,550 --> 00:19:15,330 So the dynamics of this are still very unfavorable to the small miner who wants 310 00:19:15,330 --> 00:19:17,820 to go online, order an ASIC, and start making money. 311 00:19:17,820 --> 00:19:23,600 And in fact, in almost all cases, people who have placed orders for mining hardware 312 00:19:23,600 --> 00:19:26,550 should have lost money based on the calculation that they made at the time. 313 00:19:28,070 --> 00:19:31,140 Except for one thing, which is that the price of BitCoin 314 00:19:32,540 --> 00:19:35,650 has been rising for most of the history of BitCoin. 315 00:19:35,650 --> 00:19:39,760 It has leveled off a lot in the last six months or so. 316 00:19:39,760 --> 00:19:44,117 But during the period when prices were rising, the rising prices often bailed out 317 00:19:44,117 --> 00:19:47,543 miners who would have lost money if prices had stayed constant. 318 00:19:49,725 --> 00:19:53,266 So in effect buying BitCoin ASICs has been an expensive and 319 00:19:53,266 --> 00:19:58,250 convoluted way to simply bet that the price of BitCoin would rise. 320 00:19:58,250 --> 00:20:01,200 And a lot of miners even though they've made money mining BitCoins 321 00:20:01,200 --> 00:20:04,580 would have been better off if they had just taken the money that they were going 322 00:20:04,580 --> 00:20:08,700 to spend on mining equipment invested it in BitCoins, 323 00:20:08,700 --> 00:20:12,270 held them while they appreciated in value, and then sold the BitCoins at the end. 324 00:20:15,020 --> 00:20:17,670 And now we're firmly in the era of professional mining. 325 00:20:19,080 --> 00:20:21,020 So the details there are often pretty scant, 326 00:20:21,020 --> 00:20:25,830 because companies doing this don't want to share exactly what their setup is. 327 00:20:25,830 --> 00:20:28,830 But it is known that there are some professional mining centers popping up 328 00:20:28,830 --> 00:20:30,750 around the world now. 329 00:20:30,750 --> 00:20:32,970 And here's just one picture that's been made available, 330 00:20:32,970 --> 00:20:36,380 one that came online in the last couple of months in the Republic of Georgia. 331 00:20:38,760 --> 00:20:42,770 So if you want to open a professional BitCoin mining center, 332 00:20:42,770 --> 00:20:44,000 where should you go? 333 00:20:44,000 --> 00:20:45,760 You basically need three things. 334 00:20:45,760 --> 00:20:47,840 You need cheap electricity. 335 00:20:47,840 --> 00:20:50,050 You need good network connectivity so 336 00:20:50,050 --> 00:20:53,410 that you can hear about new boxes they're announced and not miss out. 337 00:20:54,730 --> 00:20:59,096 Ideally, you like a cool climate so that you don't have to pay too much in your 338 00:20:59,096 --> 00:21:01,798 cooling bills to cool all this equipment down. 339 00:21:01,798 --> 00:21:05,790 In addition to Georgia, places like Iceland have been popular destinations for 340 00:21:05,790 --> 00:21:08,909 people to try to [COUGH] start their BitCoin mining data center. 341 00:21:12,540 --> 00:21:16,010 So if we zoom out a little bit, and we think about the evolution of mining, 342 00:21:16,010 --> 00:21:19,590 we can see really interesting parallels between Bitcoin mining and 343 00:21:19,590 --> 00:21:23,070 gold mining, or really any other kind of mining, but 344 00:21:23,070 --> 00:21:27,350 especially gold mining because it's led to the same kind of gold rush mentality, 345 00:21:27,350 --> 00:21:32,082 when initially a lot of young, amateur folks wanted to get into the business. 346 00:21:32,082 --> 00:21:37,370 So, whereas with Bitcoin mining, we've seen this slow evolution from CPUs, 347 00:21:37,370 --> 00:21:42,340 to GPUs, to FPGAs, to now ASICs, with gold mining we saw the evolution 348 00:21:42,340 --> 00:21:47,000 form individual people with a gold pan, to maybe a small team of people with a sluice 349 00:21:47,000 --> 00:21:51,750 box, to placer mining which was a big group of people 350 00:21:51,750 --> 00:21:56,080 blowing away hillsides with water, to modern gold mining 351 00:21:56,080 --> 00:22:00,730 which is a giant open pit extracting tons and tons of raw materials from the Earth. 352 00:22:02,600 --> 00:22:03,800 And in both cases, 353 00:22:03,800 --> 00:22:08,510 the friendliness to small people trying to jump into this has gone down, and there's 354 00:22:08,510 --> 00:22:12,030 been a consolidation with large companies owning most of the action over time. 355 00:22:15,560 --> 00:22:16,740 So a couple of questions for 356 00:22:16,740 --> 00:22:21,300 the future are, are small miners out of BitCoin mining forever? 357 00:22:21,300 --> 00:22:23,820 Is there any way to make it as a small miner in this game? 358 00:22:26,780 --> 00:22:29,320 And does that violate the original spirit of BitCoin? 359 00:22:29,320 --> 00:22:33,240 Does the existence of these ASICs and these large mining centers go against 360 00:22:33,240 --> 00:22:36,830 the original vision of Satoshi Nakamoto, which was to have 361 00:22:36,830 --> 00:22:40,500 every individual in the network being a miner, running on their own computer? 362 00:22:43,090 --> 00:22:47,270 So some people who think that this has been a violation of the original vision 363 00:22:47,270 --> 00:22:50,500 wonder wouldn't it be better off if there were no mining ASICs, 364 00:22:50,500 --> 00:22:55,560 if the only way to mine BitCoin was using your CPU like in the good old days. 365 00:22:58,050 --> 00:23:00,620 Well we're going to have a lecture later on in this series, 366 00:23:00,620 --> 00:23:04,320 which I'll point forward to a number of times in this lecture, 367 00:23:04,320 --> 00:23:08,110 when we look at alternative formulations of mining that might be possible. 368 00:23:08,110 --> 00:23:11,195 So I won't say anything more about how we could design mining to be less 369 00:23:11,195 --> 00:23:12,503 friendly for ASICs here, but 370 00:23:12,503 --> 00:23:16,131 that will be a topic that will come up in the future that will be quite interesting.