1 00:00:04,480 --> 00:00:12,400 Speaker 1: Welcome to tech Stuff, a production from iHeartRadio. Hey there, 2 00:00:12,400 --> 00:00:15,880 Speaker 1: and welcome to tech Stuff. I'm your host, Jonathan Strickland. 3 00:00:15,920 --> 00:00:19,439 Speaker 1: I'm an executive producer with iHeart Podcasts. And how the 4 00:00:19,520 --> 00:00:22,600 Speaker 1: tech are you? Now? There's a pretty good chance that 5 00:00:22,800 --> 00:00:27,880 Speaker 1: you've heard or read something about AI chips. But what 6 00:00:27,920 --> 00:00:31,960 Speaker 1: the heck is an AI chip? Is it a microchip 7 00:00:32,000 --> 00:00:36,720 Speaker 1: that actually has artificial intelligence incorporated directly into the semiconductor 8 00:00:36,800 --> 00:00:41,040 Speaker 1: material somehow? And if so, what does that mean? I 9 00:00:41,040 --> 00:00:42,920 Speaker 1: figured it would be a good idea to talk about 10 00:00:43,159 --> 00:00:48,159 Speaker 1: microchips and processors and AI enabled chips in particular to 11 00:00:48,240 --> 00:00:53,000 Speaker 1: help demystify everything because part of the problem, I think 12 00:00:53,200 --> 00:00:55,880 Speaker 1: is that AI chips are are kind of becoming a 13 00:00:55,920 --> 00:01:00,360 Speaker 1: marketing term. It's not just a way to describe technology. 14 00:01:00,440 --> 00:01:03,120 Speaker 1: It's a way to try and set aside a product 15 00:01:03,400 --> 00:01:06,680 Speaker 1: to try and you know, pose it as the new hotness. 16 00:01:07,000 --> 00:01:10,760 Speaker 1: And yes, I know I'm ancient and I use outdated slang. 17 00:01:11,200 --> 00:01:14,600 Speaker 1: So first, when I talk about microchips, I'm talking about 18 00:01:14,840 --> 00:01:19,600 Speaker 1: integrated circuits. Jack Kilby invented the first integrated circuit back 19 00:01:19,640 --> 00:01:24,240 Speaker 1: in nineteen fifty eight at Texas Instruments, and an integrated 20 00:01:24,280 --> 00:01:29,000 Speaker 1: circuit is a collection of interconnected electronic components that happens 21 00:01:29,040 --> 00:01:32,959 Speaker 1: to be built on top of a semiconductor material. Semiconductors, 22 00:01:32,959 --> 00:01:36,959 Speaker 1: as the name suggests, our materials that under certain conditions 23 00:01:37,000 --> 00:01:41,760 Speaker 1: will conduct electricity and under other conditions will insulate or 24 00:01:41,920 --> 00:01:45,959 Speaker 1: block the flow of electricity. The invention of the transistor, 25 00:01:46,280 --> 00:01:51,080 Speaker 1: in addition to the integrated circuit is what allowed for miniaturization. 26 00:01:51,520 --> 00:01:54,840 Speaker 1: That's why computers no longer have to be huge, you know. 27 00:01:54,880 --> 00:01:58,200 Speaker 1: I'm talking about like those old computers, those mainframes that 28 00:01:58,240 --> 00:02:01,520 Speaker 1: would fill up entire rooms or even an entire floor 29 00:02:01,600 --> 00:02:05,360 Speaker 1: of a building. Miniaturization would eventually allow for the production 30 00:02:05,480 --> 00:02:09,000 Speaker 1: of powerful personal computers that were a fraction of the 31 00:02:09,080 --> 00:02:12,640 Speaker 1: size of their predecessors, but just as powerful or even 32 00:02:12,680 --> 00:02:18,639 Speaker 1: more so. The development of arithmetic logic units or ALUs, which, 33 00:02:18,639 --> 00:02:22,640 Speaker 1: as the name suggests, are circuits designed to perform arithmetic 34 00:02:22,919 --> 00:02:28,600 Speaker 1: or mathematical functions on inputs and then produce the relevant outputs. Those, 35 00:02:28,919 --> 00:02:32,959 Speaker 1: in turn served as a building block for the development 36 00:02:33,200 --> 00:02:39,400 Speaker 1: of central processing units or CPUs. The first CPU microprocessor, 37 00:02:39,919 --> 00:02:45,200 Speaker 1: arguably was Intel's for zero zero four computer on a CHEP. 38 00:02:45,480 --> 00:02:49,360 Speaker 1: So this was a fairly limited processor, particularly if we 39 00:02:49,480 --> 00:02:52,000 Speaker 1: judge it by today's standards. I think it had like 40 00:02:52,040 --> 00:02:56,440 Speaker 1: a four bitwidth bus that would allow for processing data, 41 00:02:56,480 --> 00:03:00,240 Speaker 1: which means it could not handle very large values. But 42 00:03:00,600 --> 00:03:04,640 Speaker 1: you have to start somewhere, and the four zero zero 43 00:03:04,760 --> 00:03:08,359 Speaker 1: four was a stepping stone to Intel's eight zero zero 44 00:03:08,520 --> 00:03:12,280 Speaker 1: eight processor. That was the processor that was found in 45 00:03:12,320 --> 00:03:15,600 Speaker 1: a lot of the first commercial personal computers. They used 46 00:03:15,880 --> 00:03:20,600 Speaker 1: the eight zero zero eight processor as their CPU. Now, 47 00:03:20,680 --> 00:03:25,520 Speaker 1: a central processing unit's job is more complex than that 48 00:03:25,840 --> 00:03:30,840 Speaker 1: of an ALU. In fact, ALUs are part of a CPU. 49 00:03:30,960 --> 00:03:34,560 Speaker 1: They are a component that make up part of a CPU. 50 00:03:35,040 --> 00:03:40,160 Speaker 1: The CPU's job is to accept incoming instructions from programs, 51 00:03:40,200 --> 00:03:45,720 Speaker 1: to retrieve those instructions, to execute those instructions on data, 52 00:03:46,080 --> 00:03:50,800 Speaker 1: and to produce the relevant outcomes. And they carry out 53 00:03:50,840 --> 00:03:54,760 Speaker 1: logic operations. They send results to the appropriate destination. That 54 00:03:54,800 --> 00:03:59,960 Speaker 1: destination might be feeding back into software to continue that process, 55 00:04:00,320 --> 00:04:02,960 Speaker 1: or it might mean that you're feeding output to some 56 00:04:03,120 --> 00:04:06,480 Speaker 1: sort of output device like a display or a printer 57 00:04:06,720 --> 00:04:11,200 Speaker 1: or something along those lines. CPUs have two very broad 58 00:04:11,680 --> 00:04:15,720 Speaker 1: categories of operations. Again, this is super super high level. 59 00:04:15,760 --> 00:04:18,880 Speaker 1: I mean, we could get far more complicated than this, 60 00:04:19,200 --> 00:04:23,760 Speaker 1: but they have logic functions and memory functions. Memory being 61 00:04:24,040 --> 00:04:27,000 Speaker 1: that's where you store information so that you can reference 62 00:04:27,040 --> 00:04:30,280 Speaker 1: it quickly in order to carry out these operations, and 63 00:04:30,360 --> 00:04:33,840 Speaker 1: logic being the actual logic gates that end up defining 64 00:04:33,960 --> 00:04:37,800 Speaker 1: how data gets processed. Those are the two big components, 65 00:04:38,120 --> 00:04:40,640 Speaker 1: and there are a few different ways that we measure 66 00:04:40,720 --> 00:04:44,159 Speaker 1: CPU performance. One is we measure it by clock speed. 67 00:04:44,560 --> 00:04:46,880 Speaker 1: You can think of this as the number of instructions 68 00:04:46,920 --> 00:04:50,560 Speaker 1: the CPU is able to handle every second. So the 69 00:04:50,680 --> 00:04:54,240 Speaker 1: higher the clock speed, the more instructions the CPU is 70 00:04:54,279 --> 00:04:57,080 Speaker 1: able to handle per second. That like three point six 71 00:04:57,120 --> 00:05:01,400 Speaker 1: gigahertz would mean three point six billion operations or instructions 72 00:05:01,400 --> 00:05:05,000 Speaker 1: I should say per second. You can also have operations 73 00:05:05,000 --> 00:05:08,440 Speaker 1: that have multiple sets of instructions, so it's a little 74 00:05:08,480 --> 00:05:11,240 Speaker 1: more complicated than just saying, oh, it can handle this 75 00:05:11,320 --> 00:05:14,200 Speaker 1: number of operations per second. Now, you can also have 76 00:05:14,279 --> 00:05:18,360 Speaker 1: CPUs that have multiple cores, and a core is essentially 77 00:05:18,839 --> 00:05:23,360 Speaker 1: all the little individual components of a CPU compartmentalized so 78 00:05:23,400 --> 00:05:27,719 Speaker 1: that you have almost like multiple CPUs on a single chip. 79 00:05:27,960 --> 00:05:32,680 Speaker 1: A single core processor is like a really fast processor. 80 00:05:33,040 --> 00:05:37,640 Speaker 1: A multicore processor is one that divides the processor capabilities 81 00:05:37,680 --> 00:05:40,400 Speaker 1: into these individual cores, and you might wonder, well, why 82 00:05:40,400 --> 00:05:42,040 Speaker 1: would you want to do that, Why would you want 83 00:05:42,080 --> 00:05:45,159 Speaker 1: to take something that is typically very powerful and very 84 00:05:45,200 --> 00:05:49,320 Speaker 1: fast and then divide that up into smaller units. Well, 85 00:05:49,320 --> 00:05:53,920 Speaker 1: that's because some computational processes are able to be performed 86 00:05:53,960 --> 00:05:58,159 Speaker 1: in parallel. This means you can divide up a task 87 00:05:58,560 --> 00:06:02,680 Speaker 1: into smaller jobs and then assign those smaller jobs to 88 00:06:02,760 --> 00:06:06,360 Speaker 1: individual cores. So for these kinds of processes, a multi 89 00:06:06,440 --> 00:06:11,320 Speaker 1: core processor can sometimes be faster than a more powerful 90 00:06:11,440 --> 00:06:14,720 Speaker 1: single core processor would, And that means it's time to 91 00:06:14,800 --> 00:06:17,920 Speaker 1: use an analogy. I bust out every time I talk 92 00:06:17,960 --> 00:06:21,240 Speaker 1: about parallel processing. Fans of tech stuff who have been 93 00:06:21,279 --> 00:06:23,520 Speaker 1: around for years, you all know what's going to happen. 94 00:06:23,560 --> 00:06:26,120 Speaker 1: Go ahead and make yourself a cup of coffee or something. 95 00:06:26,320 --> 00:06:30,080 Speaker 1: So imagine you have a math class. You're a teacher. 96 00:06:30,320 --> 00:06:33,080 Speaker 1: You've got a math class, and your math class has 97 00:06:33,440 --> 00:06:36,480 Speaker 1: five students in it. It's very small focus group. One 98 00:06:36,520 --> 00:06:40,520 Speaker 1: of those five students is like a super math genius. 99 00:06:40,760 --> 00:06:43,480 Speaker 1: They are leagues ahead of the other students. The other 100 00:06:43,560 --> 00:06:47,640 Speaker 1: four students are good at math, they're great students, but 101 00:06:47,880 --> 00:06:51,040 Speaker 1: they take a little more time than the genius does 102 00:06:51,120 --> 00:06:55,080 Speaker 1: to work out Your typical math problem. So you decide 103 00:06:55,120 --> 00:06:58,080 Speaker 1: you're going to give a pop quiz to your class. 104 00:06:58,200 --> 00:07:01,480 Speaker 1: But this pop quiz is a race. It's a race 105 00:07:01,520 --> 00:07:06,960 Speaker 1: that is pitting the super genius against the other four students. Now, 106 00:07:07,279 --> 00:07:11,440 Speaker 1: if that pop quiz consisted of just one mathematical problem, 107 00:07:12,000 --> 00:07:14,880 Speaker 1: or if it had a series of math problems, but 108 00:07:14,920 --> 00:07:18,440 Speaker 1: those math problems were sequential, which means like the information 109 00:07:18,520 --> 00:07:21,240 Speaker 1: you need to solve question two can be found in 110 00:07:21,280 --> 00:07:24,360 Speaker 1: the answer of question one. If that were the case, 111 00:07:24,760 --> 00:07:28,400 Speaker 1: your super genius is gonna win, right because they would 112 00:07:28,440 --> 00:07:31,680 Speaker 1: be able to solve the problem or series of problems 113 00:07:31,800 --> 00:07:35,239 Speaker 1: much more quickly than anyone in the rest of the class. 114 00:07:35,480 --> 00:07:37,760 Speaker 1: And you can't divide that problem up. If it's a 115 00:07:37,800 --> 00:07:40,480 Speaker 1: series that you know question two depends on the outcome 116 00:07:40,480 --> 00:07:43,040 Speaker 1: of question one, you can't divide that up because you 117 00:07:43,040 --> 00:07:45,400 Speaker 1: wouldn't have the information you need to work on the 118 00:07:45,440 --> 00:07:49,800 Speaker 1: problem until the first part was solved. However, let's say 119 00:07:49,840 --> 00:07:54,200 Speaker 1: instead you make a pop quiz that has four math 120 00:07:54,240 --> 00:07:57,600 Speaker 1: problems on it. Each of these four math problems is 121 00:07:57,640 --> 00:08:00,600 Speaker 1: self contained. They do not depend upon the outcomes of 122 00:08:00,680 --> 00:08:03,480 Speaker 1: any of the other questions. So the super genius needs 123 00:08:03,520 --> 00:08:07,680 Speaker 1: to finish all four problems. But for your other four students, 124 00:08:07,840 --> 00:08:11,120 Speaker 1: they're given the option that they can each tackle a 125 00:08:11,280 --> 00:08:14,480 Speaker 1: different problem on the quiz, and if all four of 126 00:08:14,520 --> 00:08:19,480 Speaker 1: them finish whatever respective problem they've chosen first, then as 127 00:08:19,520 --> 00:08:22,720 Speaker 1: a group they win. Now, in that case, the four 128 00:08:22,760 --> 00:08:25,160 Speaker 1: students are far more likely to come out on top. 129 00:08:25,360 --> 00:08:28,360 Speaker 1: The super genius could be as far as like question 130 00:08:28,480 --> 00:08:31,320 Speaker 1: three or four, But each of the other students only 131 00:08:31,360 --> 00:08:34,160 Speaker 1: has to solve a single problem in order to complete 132 00:08:34,200 --> 00:08:38,080 Speaker 1: the pop quiz. That's like a multi core processor working 133 00:08:38,080 --> 00:08:43,439 Speaker 1: on a parallel processing problem. For some subsets of computational operations, 134 00:08:43,440 --> 00:08:46,280 Speaker 1: having multiple cores to work on things all at the 135 00:08:46,360 --> 00:08:49,240 Speaker 1: same time is a big advantage, all right. So that's 136 00:08:49,280 --> 00:08:53,800 Speaker 1: a super high level look at CPUs. Now let's turn 137 00:08:53,880 --> 00:08:58,880 Speaker 1: to GPUs. These are graphics processing units. The name actually 138 00:08:58,880 --> 00:09:02,520 Speaker 1: comes from the g Force two fifty six graphics card 139 00:09:02,559 --> 00:09:06,600 Speaker 1: from Nvidia. So in the nineteen nineties we saw the 140 00:09:06,640 --> 00:09:11,920 Speaker 1: introduction of new graphics intensive applications, particularly in things like 141 00:09:12,080 --> 00:09:16,439 Speaker 1: video editing or in video games, and the CPUs of 142 00:09:16,480 --> 00:09:20,920 Speaker 1: that era were not necessarily optimized to get the job done, 143 00:09:20,960 --> 00:09:24,120 Speaker 1: like it was more work than the CPU could typically handle. 144 00:09:24,480 --> 00:09:29,400 Speaker 1: So the performance of these kinds of programs would be substandards. 145 00:09:29,400 --> 00:09:32,320 Speaker 1: Sometimes the programs wouldn't even run on a computer that 146 00:09:32,480 --> 00:09:35,600 Speaker 1: just had a CPU, even a good CPU. So then 147 00:09:35,640 --> 00:09:39,239 Speaker 1: you had companies like in Video that began to introduce 148 00:09:39,400 --> 00:09:43,200 Speaker 1: graphics cards, and these graphics cards had integrated circuits that 149 00:09:43,240 --> 00:09:48,960 Speaker 1: were better designed. They were optimized to handle graphics processing specifically, 150 00:09:49,240 --> 00:09:52,559 Speaker 1: so that would let the CPU offload the graphics processing 151 00:09:52,679 --> 00:09:55,800 Speaker 1: jobs to the graphics card. The CPU could then focus 152 00:09:55,840 --> 00:09:59,840 Speaker 1: on other operations. The g Force two fifty six introduced 153 00:09:59,840 --> 00:10:03,440 Speaker 1: a ton of new capabilities and features. And while the 154 00:10:03,480 --> 00:10:06,760 Speaker 1: graphics processing unit name might have just started off as 155 00:10:06,840 --> 00:10:09,920 Speaker 1: kind of a marketing strategy, you know, Nvidia gave the 156 00:10:09,920 --> 00:10:12,840 Speaker 1: g Force two fifty six this designation of graphics processing 157 00:10:12,920 --> 00:10:15,280 Speaker 1: unit to kind of set it apart from other graphics 158 00:10:15,320 --> 00:10:17,600 Speaker 1: cards that were on the market. Well, it would turn 159 00:10:17,640 --> 00:10:21,920 Speaker 1: out that the GPU name would have staying power, and 160 00:10:21,960 --> 00:10:26,080 Speaker 1: today any self respecting gamer has a powerful GPU in 161 00:10:26,160 --> 00:10:29,200 Speaker 1: their gaming rig. The GPUs would grow to be more 162 00:10:29,240 --> 00:10:32,640 Speaker 1: important than CPUs, at least for some people. Though it 163 00:10:32,640 --> 00:10:35,560 Speaker 1: would be reductive to say that gamers only need a 164 00:10:35,679 --> 00:10:38,200 Speaker 1: powerful GPU and they don't have to worry about the 165 00:10:38,240 --> 00:10:40,960 Speaker 1: CPU at all. It honestly depends a lot on the 166 00:10:41,000 --> 00:10:44,360 Speaker 1: types of games they play. That is a big component. 167 00:10:44,440 --> 00:10:47,440 Speaker 1: Sometimes having a really fast GPU isn't going to help 168 00:10:47,440 --> 00:10:49,600 Speaker 1: you out that much. It all depends on the types 169 00:10:49,640 --> 00:10:51,880 Speaker 1: of processing you're doing. If you're not doing a lot 170 00:10:51,880 --> 00:10:56,200 Speaker 1: of parallel processing, then a really fast GPU isn't likely 171 00:10:56,280 --> 00:10:59,440 Speaker 1: to boost your performance that much. But the real purpose 172 00:10:59,480 --> 00:11:02,760 Speaker 1: of a GPU is to perform certain types of computational 173 00:11:02,800 --> 00:11:06,920 Speaker 1: operations very quickly and efficiently, in order to do stuff 174 00:11:06,960 --> 00:11:11,680 Speaker 1: like speed up image creation, video and animation. As it 175 00:11:11,679 --> 00:11:15,160 Speaker 1: would turn out, GPUs would also be handy for other things. 176 00:11:15,440 --> 00:11:20,040 Speaker 1: So your typical GPU consists of many specialized processor cores. 177 00:11:20,360 --> 00:11:23,200 Speaker 1: These cores are not designed to do everything you know. 178 00:11:23,240 --> 00:11:26,360 Speaker 1: They do a subset of operations really well, but if 179 00:11:26,360 --> 00:11:29,000 Speaker 1: you ask a GPU to do something outside of that, 180 00:11:29,080 --> 00:11:32,000 Speaker 1: it's not going to perform at you know, at the 181 00:11:32,000 --> 00:11:35,000 Speaker 1: same level as your typical CPU would. But this does 182 00:11:35,080 --> 00:11:38,800 Speaker 1: mean a GPU is a fantastic tool for specific operations 183 00:11:39,040 --> 00:11:42,560 Speaker 1: and then less useful for others. Apart from processing graphics, 184 00:11:42,600 --> 00:11:45,400 Speaker 1: GPUs have been found to be really useful in applications 185 00:11:45,480 --> 00:11:50,000 Speaker 1: ranging from machine learning projects to proof of work cryptocurrency 186 00:11:50,320 --> 00:11:55,720 Speaker 1: mining operations. Now to be clear, GPUs, at least until recently, 187 00:11:55,960 --> 00:12:00,000 Speaker 1: occupied a kind of a sweet space in crypto minds. 188 00:12:00,679 --> 00:12:02,760 Speaker 1: They are not the top of the heap when it 189 00:12:02,800 --> 00:12:07,120 Speaker 1: comes to crypto mining integrated circuits. We'll get to the 190 00:12:07,200 --> 00:12:10,400 Speaker 1: kind that are used in high end crypto mining in 191 00:12:10,520 --> 00:12:13,280 Speaker 1: just a little bit. So for stuff like Bitcoin, which 192 00:12:13,320 --> 00:12:15,920 Speaker 1: as I record this episode, is trading at around fifty 193 00:12:15,920 --> 00:12:18,480 Speaker 1: eight thousand dollars per coin. In fact, I think it's 194 00:12:18,480 --> 00:12:23,000 Speaker 1: like fifty eight point five thousand. That's a lot of money. Well, 195 00:12:23,160 --> 00:12:25,720 Speaker 1: if you're using GPUs, you're not going to cut it. 196 00:12:25,920 --> 00:12:28,520 Speaker 1: You're not going to compete in that space. GPUs just 197 00:12:28,600 --> 00:12:33,080 Speaker 1: can't operate at a level that would make it feasible 198 00:12:33,120 --> 00:12:36,240 Speaker 1: for you to use them for your mining operations. That's 199 00:12:36,280 --> 00:12:40,240 Speaker 1: because the value of bitcoin is so high that it 200 00:12:40,400 --> 00:12:45,360 Speaker 1: drives cryptocurrency miners to seek out the absolute top tier processors, 201 00:12:45,600 --> 00:12:50,160 Speaker 1: and GPUs, while they're great, they're really more mid tier. 202 00:12:50,760 --> 00:12:52,880 Speaker 1: Now it helps if you know what proof of work 203 00:12:53,000 --> 00:12:57,239 Speaker 1: crypto mining is all about. So with proof of work systems, 204 00:12:57,520 --> 00:13:00,040 Speaker 1: you have a network of machines that make up this 205 00:13:00,360 --> 00:13:04,120 Speaker 1: cryptocurrency network, such as bitcoin. We'll use Bitcoin as the 206 00:13:04,200 --> 00:13:07,640 Speaker 1: main example because that was sort of the progenitor of 207 00:13:07,679 --> 00:13:12,360 Speaker 1: this space. So every so often the network issues a challenge, 208 00:13:12,360 --> 00:13:15,600 Speaker 1: which is to solve a mathematical problem, and if you 209 00:13:15,679 --> 00:13:18,000 Speaker 1: do solve it, if you're the first one to solve it, 210 00:13:18,080 --> 00:13:21,120 Speaker 1: you will receive some crypto coins as a reward. The 211 00:13:21,160 --> 00:13:24,440 Speaker 1: act of solving typically is tied to validating a block 212 00:13:24,520 --> 00:13:29,200 Speaker 1: of crypto transactions, So the problem's complexity will depend upon 213 00:13:29,280 --> 00:13:33,400 Speaker 1: a couple of different things. Typically, there's an ideal amount 214 00:13:33,440 --> 00:13:37,840 Speaker 1: of time that it should take to solve this mathematical problem. 215 00:13:38,040 --> 00:13:41,720 Speaker 1: For bitcoin, that time is ten minutes. The other thing 216 00:13:41,760 --> 00:13:44,440 Speaker 1: that determines the complexity of the problem is how much 217 00:13:44,480 --> 00:13:48,000 Speaker 1: computing power is being thrown at solving the problem in 218 00:13:48,040 --> 00:13:51,160 Speaker 1: the first place. So let's go back to our classroom analogy. 219 00:13:51,280 --> 00:13:54,640 Speaker 1: Let's say that you're creating a test, and for whatever reason, 220 00:13:54,720 --> 00:13:57,720 Speaker 1: you have decided this test should take the students ten 221 00:13:57,800 --> 00:14:01,720 Speaker 1: minutes to complete, so you're not really focused on any 222 00:14:01,760 --> 00:14:04,160 Speaker 1: other outcome other than trying to make a test that's 223 00:14:04,200 --> 00:14:08,559 Speaker 1: going to take ten minutes to complete. However, you've misjudged 224 00:14:08,559 --> 00:14:11,960 Speaker 1: the difficulty. Maybe one of your students hands in their 225 00:14:12,040 --> 00:14:14,920 Speaker 1: test six minutes in. Now you know you need to 226 00:14:14,960 --> 00:14:17,400 Speaker 1: make the next test harder in order to hit this 227 00:14:17,440 --> 00:14:21,240 Speaker 1: seemingly arbitrary goal of ten minutes. On the flip side, 228 00:14:21,400 --> 00:14:23,560 Speaker 1: let's say the first student to solve the test took 229 00:14:23,640 --> 00:14:26,680 Speaker 1: fifteen minutes to complete it. Then you know your test 230 00:14:26,720 --> 00:14:29,240 Speaker 1: is too hard and you need to ease up a 231 00:14:29,240 --> 00:14:32,040 Speaker 1: little bit for the next test. When the value of 232 00:14:32,080 --> 00:14:35,520 Speaker 1: cryptocurrency goes up, there's a greater incentive to be the 233 00:14:35,600 --> 00:14:39,440 Speaker 1: first to solve the mathematical problem because the reward is larger. 234 00:14:39,800 --> 00:14:43,840 Speaker 1: That drives miners to buy more processors and to network 235 00:14:43,880 --> 00:14:47,320 Speaker 1: them together, and these are processors that are particularly good 236 00:14:47,360 --> 00:14:50,480 Speaker 1: at solving the types of problems that you get when 237 00:14:50,480 --> 00:14:55,000 Speaker 1: you're crypto mining. For a while, that meant GPUs they 238 00:14:55,000 --> 00:14:58,760 Speaker 1: were the best. But the value of bitcoin went up 239 00:14:58,840 --> 00:15:03,160 Speaker 1: and up and up, and there were other options besides GPUs. 240 00:15:03,240 --> 00:15:06,400 Speaker 1: There were options that were more expensive than GPUs, so 241 00:15:06,440 --> 00:15:09,480 Speaker 1: it require a bigger investment, but then on top of that, 242 00:15:09,520 --> 00:15:11,800 Speaker 1: you were looking at bigger rewards, so it made that 243 00:15:11,840 --> 00:15:17,120 Speaker 1: investment worthwhile. So the integrated circuit that typically replaces GPUs 244 00:15:17,160 --> 00:15:21,080 Speaker 1: for high end cryptocurrency mining, those would be application specific 245 00:15:21,160 --> 00:15:25,200 Speaker 1: integrated circuits or AASAC ASIC. We'll get to those in 246 00:15:25,520 --> 00:15:28,920 Speaker 1: just a bit, so you could if you wanted to 247 00:15:29,080 --> 00:15:32,920 Speaker 1: still run mining rigs using GPUs, nothing would stop you 248 00:15:33,120 --> 00:15:35,560 Speaker 1: from doing that, but you'd be going up against people 249 00:15:35,600 --> 00:15:39,960 Speaker 1: with networks and machines running much more streamlined optimized processors, 250 00:15:40,240 --> 00:15:44,040 Speaker 1: so you would be unlikely to beat them. Okay, we 251 00:15:44,160 --> 00:15:46,560 Speaker 1: got a lot more to cover, but let's take a 252 00:15:46,600 --> 00:15:58,720 Speaker 1: quick break to thank our sponsors. Okay, we're coming back 253 00:15:58,720 --> 00:16:00,920 Speaker 1: to talk a little bit more about crypt currency mining 254 00:16:00,920 --> 00:16:06,120 Speaker 1: in GPUs. So for a while, people who were crypto 255 00:16:06,240 --> 00:16:11,040 Speaker 1: mining ethereum would stick with GPUs. The reason for this 256 00:16:11,120 --> 00:16:14,920 Speaker 1: is ethereum had a lower value, much lower than bitcoin. 257 00:16:15,200 --> 00:16:17,280 Speaker 1: All right, We're talking about a few thousand dollars as 258 00:16:17,280 --> 00:16:20,880 Speaker 1: opposed to tens of thousands of dollars per coin, and 259 00:16:20,920 --> 00:16:26,240 Speaker 1: this meant that it would be impractical to use high 260 00:16:26,400 --> 00:16:31,000 Speaker 1: end integrade circuits like AASC circuits for mining ethereum because 261 00:16:31,040 --> 00:16:34,680 Speaker 1: the cost of doing so would be such that you 262 00:16:34,680 --> 00:16:37,760 Speaker 1: wouldn't be making up that cost in the profit you 263 00:16:38,280 --> 00:16:43,040 Speaker 1: gained from mining the cryptocurrency. So sticking with GPUs made 264 00:16:43,080 --> 00:16:47,120 Speaker 1: more sense, right because from an economic standpoint, that was 265 00:16:47,680 --> 00:16:52,560 Speaker 1: the sweet spot. However, then Ethereum switched to proof of 266 00:16:52,760 --> 00:16:56,360 Speaker 1: steak instead of proof of work. Proof of steak doesn't 267 00:16:56,360 --> 00:16:59,400 Speaker 1: do that whole math problem thing at all, and the 268 00:16:59,440 --> 00:17:02,960 Speaker 1: demand for GPUs and crypto mining plummeted as a result. 269 00:17:03,200 --> 00:17:05,679 Speaker 1: There are other cryptocurrencies out there, some of which that 270 00:17:06,440 --> 00:17:09,119 Speaker 1: still do use proof of work, but they're not as 271 00:17:09,200 --> 00:17:14,120 Speaker 1: sought after as Bitcoin or ethereum are. So this meant 272 00:17:14,200 --> 00:17:17,439 Speaker 1: that the demand for GPUs in the crypto space began 273 00:17:17,560 --> 00:17:20,640 Speaker 1: to diminish, and that became really good news for people 274 00:17:20,680 --> 00:17:23,679 Speaker 1: who wanted a GPU for something else, like for a 275 00:17:23,680 --> 00:17:27,280 Speaker 1: gaming rig for example. Now, I would say for the 276 00:17:27,359 --> 00:17:31,640 Speaker 1: majority of people out there, like your average consumers, CPUs 277 00:17:31,720 --> 00:17:34,680 Speaker 1: and GPUs are the beginning and end of it when 278 00:17:34,720 --> 00:17:38,159 Speaker 1: it comes to processors or microchips that are meant to 279 00:17:38,280 --> 00:17:41,280 Speaker 1: act like processors, But there are a couple of other 280 00:17:41,359 --> 00:17:45,080 Speaker 1: varieties out there that we use for special purposes. And 281 00:17:45,119 --> 00:17:48,880 Speaker 1: the special purpose thing is the important part to keep 282 00:17:48,920 --> 00:17:52,400 Speaker 1: in mind. A CPU, by necessity, has to be able 283 00:17:52,400 --> 00:17:55,200 Speaker 1: to do a bit of everything right. Because a CPU 284 00:17:55,600 --> 00:18:00,240 Speaker 1: is the control center of your typical computer. It needs 285 00:18:00,280 --> 00:18:02,600 Speaker 1: to be able to handle operations from a variety of 286 00:18:02,600 --> 00:18:05,320 Speaker 1: different programs and that kind of thing. It is a 287 00:18:05,400 --> 00:18:08,080 Speaker 1: jack of all trades master of none. It needs to 288 00:18:08,119 --> 00:18:10,239 Speaker 1: be able to handle whatever you throw at it, but 289 00:18:10,280 --> 00:18:14,840 Speaker 1: that means it cannot be optimized for any specific task. 290 00:18:15,320 --> 00:18:19,760 Speaker 1: So what it lacks in efficiency, it makes up for inversatility. 291 00:18:20,240 --> 00:18:24,240 Speaker 1: GPUs are more specialized and so they can handle certain 292 00:18:24,280 --> 00:18:29,240 Speaker 1: processes better than a CPU typically can, But a GPU 293 00:18:29,400 --> 00:18:32,199 Speaker 1: might not be so good at executing all the different 294 00:18:32,240 --> 00:18:35,080 Speaker 1: tasks that a CPU has to handle, So while it 295 00:18:35,240 --> 00:18:39,520 Speaker 1: is faster with some stuff, it's slower with other stuff. Now, 296 00:18:39,560 --> 00:18:42,240 Speaker 1: the next two types of semiconductor devices I want to 297 00:18:42,280 --> 00:18:46,280 Speaker 1: mention are even more specialized than GPUs, and then we'll 298 00:18:46,400 --> 00:18:52,159 Speaker 1: end with one that is specialized specifically for the AI field. 299 00:18:52,720 --> 00:18:57,560 Speaker 1: So next up is the field programmable gait arrays or 300 00:18:57,720 --> 00:19:02,879 Speaker 1: FPGA's now a definition from XLinks dot com because x 301 00:19:02,920 --> 00:19:07,920 Speaker 1: links is what introduced this technology back in the mid 302 00:19:08,080 --> 00:19:13,479 Speaker 1: nineteen eighties. So x links defines this as FPGA's quote. 303 00:19:13,640 --> 00:19:18,480 Speaker 1: Are based around a matrix of configurable logic blocks CLBs 304 00:19:18,680 --> 00:19:23,840 Speaker 1: connected via programmable interconnects. End quote that sounds like gibberish 305 00:19:23,920 --> 00:19:29,200 Speaker 1: to some folks. It's definitely got some barriers there from 306 00:19:29,240 --> 00:19:33,040 Speaker 1: easy understanding, but the idea is pretty simple. When you 307 00:19:33,040 --> 00:19:37,000 Speaker 1: boil it down to what's basically happening. So imagine that 308 00:19:37,080 --> 00:19:40,479 Speaker 1: you have a microchip and you're able to reconfigure the 309 00:19:40,640 --> 00:19:44,360 Speaker 1: individual components on that microchip so that they're optimized for 310 00:19:44,440 --> 00:19:46,960 Speaker 1: whatever it is you need to do. So you can 311 00:19:47,080 --> 00:19:50,879 Speaker 1: reprogram this chip, in other words, so that it is 312 00:19:50,960 --> 00:19:54,880 Speaker 1: better aligned with the task you have at hand. As 313 00:19:54,920 --> 00:19:57,640 Speaker 1: I said, x links first introduced this type of integrated 314 00:19:57,640 --> 00:20:00,199 Speaker 1: circuit back in nineteen eighty five, and the aim us 315 00:20:00,240 --> 00:20:02,800 Speaker 1: to make an integrated circuit that could potentially fit the 316 00:20:02,840 --> 00:20:07,240 Speaker 1: needs of different specific use cases, not by being a 317 00:20:07,320 --> 00:20:10,639 Speaker 1: jack of all trades that could do anything, but do 318 00:20:10,800 --> 00:20:13,440 Speaker 1: so at a kind of a mediocre level, but rather 319 00:20:13,600 --> 00:20:18,840 Speaker 1: by being configured to work best for that specific application. Moreover, 320 00:20:19,560 --> 00:20:23,800 Speaker 1: you can at least sometimes reconfigure without having to change 321 00:20:23,840 --> 00:20:27,439 Speaker 1: the actual physical architecture of the chip itself. This is 322 00:20:27,480 --> 00:20:30,640 Speaker 1: important because not everyone has access to a clean room 323 00:20:30,840 --> 00:20:35,439 Speaker 1: with incredibly precise and computer operated tools. That's exactly what 324 00:20:35,480 --> 00:20:37,800 Speaker 1: you would need if you wanted to perform surgery on 325 00:20:37,880 --> 00:20:43,679 Speaker 1: a microchip. Instead, an FPGA has these CLBs that x 326 00:20:43,720 --> 00:20:47,439 Speaker 1: links talked about, the configurable logic blocks. These can be 327 00:20:47,480 --> 00:20:50,679 Speaker 1: programmed to act like simple logic gates, and these gates 328 00:20:50,680 --> 00:20:54,800 Speaker 1: follow specific rules. Essentially, they either allow electrical current to 329 00:20:54,800 --> 00:20:58,640 Speaker 1: flow through or they block it from flowing through, and this, 330 00:20:59,040 --> 00:21:01,600 Speaker 1: when you look at it a macro level, is what 331 00:21:01,800 --> 00:21:06,720 Speaker 1: allows operations on a processor. The field and field programmable 332 00:21:06,840 --> 00:21:10,560 Speaker 1: gate array means you can actually do this reprogramming after 333 00:21:10,640 --> 00:21:14,840 Speaker 1: the FPGA has shipped from its manufacturer. So instead of 334 00:21:14,960 --> 00:21:18,439 Speaker 1: working with a manufacturer to specialize a chip from the 335 00:21:18,520 --> 00:21:21,760 Speaker 1: design phase and then go all the way through to manufacturing, 336 00:21:22,119 --> 00:21:26,600 Speaker 1: the manufacturer makes this FPGA that can potentially be one 337 00:21:26,720 --> 00:21:30,159 Speaker 1: of thousands of different configurations, and then you program it 338 00:21:30,200 --> 00:21:33,639 Speaker 1: once you receive it. Now, some of these FPGAs are 339 00:21:33,680 --> 00:21:37,399 Speaker 1: limited to kind of a one time only configuration, so 340 00:21:37,600 --> 00:21:40,720 Speaker 1: you can program them once you get them, but then 341 00:21:40,760 --> 00:21:45,760 Speaker 1: they're set in that particular configuration from that point forward. 342 00:21:45,880 --> 00:21:49,200 Speaker 1: But others are designed so that they can be reprogrammed 343 00:21:49,440 --> 00:21:52,359 Speaker 1: multiple times, which obviously makes them very useful. If you 344 00:21:52,440 --> 00:21:57,399 Speaker 1: wanted to prototype a technology and you aren't really sure 345 00:21:57,520 --> 00:22:00,320 Speaker 1: which configuration is going to be best for whatever it 346 00:22:00,359 --> 00:22:02,359 Speaker 1: is you're trying to do, it's great to have a 347 00:22:02,480 --> 00:22:07,399 Speaker 1: chip you can reprogram so you can try different configurations 348 00:22:07,440 --> 00:22:09,760 Speaker 1: to find the one that makes the most sense for 349 00:22:09,880 --> 00:22:13,159 Speaker 1: whatever it is you're trying to achieve. One issue with 350 00:22:13,400 --> 00:22:16,840 Speaker 1: FPGA's is that they are not cost efficient when you're 351 00:22:16,880 --> 00:22:21,000 Speaker 1: looking at mass production. They're great if you are prototyping, 352 00:22:21,440 --> 00:22:23,760 Speaker 1: but if you plan to make a whole bunch of them, 353 00:22:23,840 --> 00:22:27,439 Speaker 1: it gets time consuming and expensive because not only do 354 00:22:27,480 --> 00:22:28,919 Speaker 1: you have to have them made, then you have to 355 00:22:28,920 --> 00:22:32,440 Speaker 1: have them programmed. Plus sometimes you may have an application 356 00:22:32,520 --> 00:22:36,200 Speaker 1: in mind that an FPGA cannot accommodate even with all 357 00:22:36,200 --> 00:22:39,639 Speaker 1: the reconfiguring. So think of an FPGA as having a 358 00:22:39,680 --> 00:22:42,760 Speaker 1: limited number of configurations and it turns out that what 359 00:22:42,840 --> 00:22:46,040 Speaker 1: you need is outside of this range. That would mean 360 00:22:46,080 --> 00:22:49,240 Speaker 1: you would need to add additional integrated circuits to your 361 00:22:49,320 --> 00:22:54,359 Speaker 1: system to accommodate these limitations of the FPGA itself, which 362 00:22:54,359 --> 00:22:57,359 Speaker 1: means you're adding more complexity to your system, and that 363 00:22:57,400 --> 00:23:01,440 Speaker 1: in turn also means you're adding more costs to your system. 364 00:23:01,680 --> 00:23:04,919 Speaker 1: Next up, we have the one I mentioned earlier, the 365 00:23:05,040 --> 00:23:11,600 Speaker 1: Application specific integrated circuit or AZC ASIC, as the name indicates, 366 00:23:11,640 --> 00:23:16,320 Speaker 1: These chips are made to operate for specific applications, and 367 00:23:16,480 --> 00:23:21,639 Speaker 1: as such, they are highly optimized from the hardware level 368 00:23:21,840 --> 00:23:25,080 Speaker 1: up for that purpose. They are not meant to be 369 00:23:25,200 --> 00:23:28,720 Speaker 1: general purpose processors like a CPU. So if you put 370 00:23:28,760 --> 00:23:31,320 Speaker 1: an ASK to work on a task that it was 371 00:23:31,480 --> 00:23:34,240 Speaker 1: not designed to handle, you are not going to get 372 00:23:34,240 --> 00:23:36,480 Speaker 1: a good result. In fact, it may not work at all. 373 00:23:36,800 --> 00:23:39,760 Speaker 1: But when it's integrated into a system that's meant to 374 00:23:39,840 --> 00:23:43,439 Speaker 1: do that one thing it was designed to do, it 375 00:23:43,520 --> 00:23:46,840 Speaker 1: does it really well. And AAK can be a speed 376 00:23:47,000 --> 00:23:51,320 Speaker 1: demon and operate at an efficiency that's much more desirable 377 00:23:51,320 --> 00:23:55,120 Speaker 1: than your typical CPU or even GPU. So unlike an 378 00:23:55,200 --> 00:24:00,920 Speaker 1: FPGA and ASK cannot be reconfigured. It is a high tech, 379 00:24:01,560 --> 00:24:05,280 Speaker 1: highly specialized chip. There are a few different approaches to 380 00:24:05,600 --> 00:24:09,320 Speaker 1: create that specialization during the manufacturing process, but I feel 381 00:24:09,359 --> 00:24:11,840 Speaker 1: like that's beyond the scope of this episode. I'll save 382 00:24:11,880 --> 00:24:14,360 Speaker 1: it for a time when I do a full episode 383 00:24:14,480 --> 00:24:19,600 Speaker 1: about AZK chips. Now, the design process for AZIK is complicated. 384 00:24:19,680 --> 00:24:22,880 Speaker 1: So imagine you're building a chip intended to do one 385 00:24:22,960 --> 00:24:25,720 Speaker 1: thing extremely well. You would have to do a lot 386 00:24:25,720 --> 00:24:27,680 Speaker 1: of work to make sure that the chip you were 387 00:24:27,720 --> 00:24:31,400 Speaker 1: designing met that purpose. So that means there's a lot 388 00:24:31,400 --> 00:24:34,520 Speaker 1: of R and D and there's a lot of testing. However, 389 00:24:34,960 --> 00:24:38,320 Speaker 1: once you do arrive at this final design, one big 390 00:24:38,359 --> 00:24:42,280 Speaker 1: advantage of AZK over FPGA is that it can then 391 00:24:42,400 --> 00:24:47,000 Speaker 1: go into large volume production. So while the development process 392 00:24:47,119 --> 00:24:50,520 Speaker 1: of an AZK is typically longer and more expensive than 393 00:24:51,040 --> 00:24:55,360 Speaker 1: using an FPGA, once you do get to the production stage, 394 00:24:55,640 --> 00:24:58,880 Speaker 1: the AZIC chips become more cost effective. So if you're 395 00:24:58,880 --> 00:25:02,400 Speaker 1: doing a one off, FPGA makes the most sense financially. 396 00:25:02,800 --> 00:25:05,600 Speaker 1: If your goal is to make something that you're going 397 00:25:05,680 --> 00:25:09,560 Speaker 1: to mass produce, AZIC makes far more sense. ASIC chips 398 00:25:09,560 --> 00:25:12,600 Speaker 1: also tend to be more power efficient than FPGA's, so 399 00:25:12,720 --> 00:25:16,359 Speaker 1: by their nature, an FPGA needs to have components that 400 00:25:16,400 --> 00:25:21,439 Speaker 1: aren't necessary for all applications because the whole point of 401 00:25:21,440 --> 00:25:25,040 Speaker 1: an FPGA is that you can reprogram them to do 402 00:25:25,080 --> 00:25:28,800 Speaker 1: specific tasks, but not every task is going to need 403 00:25:29,280 --> 00:25:33,120 Speaker 1: every component that's on that circuit. So that means there's 404 00:25:33,160 --> 00:25:36,200 Speaker 1: going to be some extra stuff on that integrated circuit 405 00:25:36,200 --> 00:25:40,480 Speaker 1: that ends up being superfluous for certain operations. With AZC, 406 00:25:40,840 --> 00:25:43,840 Speaker 1: you can leave off anything that would be superfluous, right, 407 00:25:43,960 --> 00:25:46,679 Speaker 1: You can leave that out of the design because you 408 00:25:46,760 --> 00:25:49,080 Speaker 1: know ahead of time what you're putting this chip to 409 00:25:49,119 --> 00:25:52,240 Speaker 1: work for, so you can only focus on the things 410 00:25:52,280 --> 00:25:55,919 Speaker 1: that are absolutely necessary for the operation of that chip. 411 00:25:56,359 --> 00:25:58,800 Speaker 1: That means you don't have to supply power to components 412 00:25:58,840 --> 00:26:02,240 Speaker 1: that aren't actually doing anything. That keeps your power consumption 413 00:26:02,359 --> 00:26:05,439 Speaker 1: costs lower in the long run. Thus, ASIC chips are 414 00:26:05,480 --> 00:26:08,040 Speaker 1: more efficient. Now, most of us are not going to 415 00:26:08,040 --> 00:26:12,000 Speaker 1: be shopping around for ASK chips. Your average consumer has 416 00:26:12,160 --> 00:26:16,440 Speaker 1: no need for them. But for folks like cryptomners, AZK 417 00:26:16,840 --> 00:26:21,800 Speaker 1: might make sense once you reach a certain level of profit. Right, 418 00:26:21,840 --> 00:26:24,399 Speaker 1: once you reach a certain level at least potential profit 419 00:26:24,840 --> 00:26:29,000 Speaker 1: if you mine a block of the cryptocurrency. Because again, 420 00:26:29,080 --> 00:26:31,719 Speaker 1: Bitcoin created the perfect storm for this back when it 421 00:26:31,920 --> 00:26:35,800 Speaker 1: was awarding six point two five coins per block, so 422 00:26:35,840 --> 00:26:39,199 Speaker 1: that meant in an average day the system would release, 423 00:26:39,280 --> 00:26:44,199 Speaker 1: or rather miners would mine around nine hundred bitcoins total 424 00:26:44,480 --> 00:26:48,680 Speaker 1: per day, and with bitcoins trading at fifty grand each, 425 00:26:49,160 --> 00:26:52,000 Speaker 1: that would mean around forty five million dollars worth of 426 00:26:52,040 --> 00:26:56,879 Speaker 1: bitcoin were up for grabs every single day. That's what 427 00:26:57,200 --> 00:27:00,439 Speaker 1: justified spending the huge amount of money it costs to 428 00:27:00,680 --> 00:27:04,320 Speaker 1: develop and deploy ACC chips for the specific task of 429 00:27:04,480 --> 00:27:09,879 Speaker 1: mining bitcoin. Yes, that design process is incredibly expensive, but 430 00:27:10,160 --> 00:27:12,439 Speaker 1: if you could create a system that could grab a 431 00:27:12,480 --> 00:27:16,480 Speaker 1: significant number of bitcoins every day, then it would pay 432 00:27:16,520 --> 00:27:19,600 Speaker 1: for itself pretty darn quickly. You might not get all 433 00:27:19,640 --> 00:27:22,280 Speaker 1: the bitcoins, you might not even get most of them, 434 00:27:22,440 --> 00:27:24,600 Speaker 1: but as long as you were grabbing a decent number 435 00:27:24,680 --> 00:27:28,440 Speaker 1: every single day, you would quickly accumulate wealth and justify 436 00:27:28,480 --> 00:27:33,000 Speaker 1: the cost of using ACAC technology. That's what left GPU 437 00:27:33,119 --> 00:27:36,679 Speaker 1: miners in the dust, because once acc systems joined the party, 438 00:27:37,200 --> 00:27:40,120 Speaker 1: the GPUs just could not compete. It would be kind 439 00:27:40,160 --> 00:27:42,680 Speaker 1: of like if you put me in the one hundred 440 00:27:42,760 --> 00:27:46,040 Speaker 1: meter dash in the Olympics, the lead runner would be 441 00:27:46,040 --> 00:27:48,080 Speaker 1: crossing the finish line before I managed to get a 442 00:27:48,160 --> 00:27:51,200 Speaker 1: quarter of the way there. Now I should add that 443 00:27:51,520 --> 00:27:54,439 Speaker 1: this year, in twenty twenty four, the number of bitcoins 444 00:27:54,520 --> 00:27:59,400 Speaker 1: awarded per block dropped by half. This was all part 445 00:27:59,400 --> 00:28:02,680 Speaker 1: of the plan. This wasn't a mistake or something. Now, 446 00:28:02,680 --> 00:28:06,359 Speaker 1: if you mine a block, instead of getting six point 447 00:28:06,400 --> 00:28:09,720 Speaker 1: twenty five coins, you end up getting three point one 448 00:28:09,880 --> 00:28:15,000 Speaker 1: two five. So again, this was this was planned, and 449 00:28:15,560 --> 00:28:18,000 Speaker 1: every four years or so the system cuts the number 450 00:28:18,000 --> 00:28:22,560 Speaker 1: of coins awarded per block mind by fifty percent. When 451 00:28:22,640 --> 00:28:25,359 Speaker 1: bitcoin first hit the scene back in early two thousand 452 00:28:25,359 --> 00:28:28,720 Speaker 1: and nine, if you mined a block successfully you would 453 00:28:28,880 --> 00:28:33,359 Speaker 1: net yourself fifty bitcoins per pop. But of course, back 454 00:28:33,359 --> 00:28:35,639 Speaker 1: in two thousand and nine, the value of bitcoin was 455 00:28:35,760 --> 00:28:40,320 Speaker 1: fractions of a cent. You wouldn't apply AASAC technology to 456 00:28:40,480 --> 00:28:43,160 Speaker 1: bitcoin mining back in those days because the coins weren't 457 00:28:43,200 --> 00:28:47,200 Speaker 1: really worth anything. In fact, on May twenty second, twenty ten, 458 00:28:47,360 --> 00:28:50,640 Speaker 1: this is a famous date in crypto history. This was 459 00:28:50,680 --> 00:28:54,760 Speaker 1: more than a year after bitcoin had launched. A cryptocurrency 460 00:28:54,840 --> 00:28:59,080 Speaker 1: minor named Laslow's spent ten thousand bitcoins in order it 461 00:28:59,120 --> 00:29:02,520 Speaker 1: to order ap pizza. So today that pizza would be 462 00:29:02,560 --> 00:29:06,280 Speaker 1: worth more than five hundred and eighty five million dollars. 463 00:29:07,040 --> 00:29:10,000 Speaker 1: And in fact, another interesting point, Bitcoin is a lot 464 00:29:10,040 --> 00:29:13,360 Speaker 1: of volatility. When I started work on this episode, it 465 00:29:13,400 --> 00:29:15,800 Speaker 1: was trading at fifty seven thousand dollars and now it's 466 00:29:15,800 --> 00:29:19,760 Speaker 1: at more than fifty eight thousand, so the value changes 467 00:29:19,800 --> 00:29:23,640 Speaker 1: pretty drastically. Anyway, getting back to the having, part of 468 00:29:23,680 --> 00:29:26,720 Speaker 1: the bitcoin strategy is that there's a finite number of 469 00:29:26,760 --> 00:29:30,080 Speaker 1: bitcoin that will ever be released into circulation, and once 470 00:29:30,320 --> 00:29:33,520 Speaker 1: the last one is in circulation, no more new bitcoin 471 00:29:33,560 --> 00:29:37,200 Speaker 1: will be minted. So specifically, that makes up twenty one 472 00:29:37,440 --> 00:29:41,640 Speaker 1: million bitcoin to control the release of bitcoin into circulation. 473 00:29:41,760 --> 00:29:46,240 Speaker 1: The system does this having business every four years, so 474 00:29:46,400 --> 00:29:49,000 Speaker 1: today mining a block on the Bitcoin network will earn 475 00:29:49,080 --> 00:29:52,120 Speaker 1: you three point one two five bitcoins, or around one 476 00:29:52,240 --> 00:29:55,480 Speaker 1: hundred and eighty one thousand dollars worth of bitcoin crypto 477 00:29:55,920 --> 00:30:01,000 Speaker 1: per block. Mind, these kinds of changes affect mining operations 478 00:30:01,040 --> 00:30:05,160 Speaker 1: because if the magic number dips too much, then it 479 00:30:05,200 --> 00:30:08,920 Speaker 1: would cost more to mine bitcoin. Then you would get 480 00:30:09,200 --> 00:30:12,280 Speaker 1: out of mining it, so you would have to adjust 481 00:30:12,320 --> 00:30:14,880 Speaker 1: your strategy. Right, you'd say, all right, well, now it 482 00:30:14,880 --> 00:30:18,920 Speaker 1: doesn't make sense for me to operate this massive computer 483 00:30:19,120 --> 00:30:24,920 Speaker 1: network of AASC machines that's drawing power directly from a 484 00:30:25,080 --> 00:30:30,240 Speaker 1: formerly decommissioned power plant because the cost of operations is 485 00:30:30,800 --> 00:30:33,840 Speaker 1: sky high and the amount that I'm able to actually 486 00:30:34,000 --> 00:30:38,400 Speaker 1: mine is much lower. Now I have one more initialism 487 00:30:38,480 --> 00:30:41,360 Speaker 1: to throw your way, but before we get to that, 488 00:30:42,000 --> 00:30:54,440 Speaker 1: let's take another quick break to thank our sponsors. Okay, 489 00:30:54,480 --> 00:30:57,120 Speaker 1: we're back, and we've talked about CPUs, and we've talked 490 00:30:57,120 --> 00:31:01,400 Speaker 1: about GPUs, and we've talked about FPGAs talked about ASICs. 491 00:31:01,840 --> 00:31:06,760 Speaker 1: Now it's time to talk about NPUs. And as a nancy, 492 00:31:07,600 --> 00:31:12,400 Speaker 1: the initialism stands for neural processing unit. These have technically 493 00:31:12,440 --> 00:31:15,800 Speaker 1: been around for a few years now, but the term 494 00:31:15,880 --> 00:31:19,160 Speaker 1: is still fairly new. For mainstream audiences. I think you 495 00:31:19,280 --> 00:31:24,080 Speaker 1: started to see them pop up in mainstream tech journals 496 00:31:24,200 --> 00:31:26,680 Speaker 1: last year, but they've been around for a few years. 497 00:31:27,160 --> 00:31:30,480 Speaker 1: And NPU is a chip with a specialized design meant 498 00:31:30,480 --> 00:31:35,080 Speaker 1: for AI applications and artificial neural networks in particular. Now, 499 00:31:35,120 --> 00:31:38,680 Speaker 1: just in case artificial neural networks, if that term is 500 00:31:38,840 --> 00:31:42,520 Speaker 1: new to you, it is a network of processors that 501 00:31:42,720 --> 00:31:47,200 Speaker 1: collectively mimics the way our neurons interconnect with one another 502 00:31:47,280 --> 00:31:50,640 Speaker 1: in our brain meet. That's a very high level and 503 00:31:50,760 --> 00:31:54,440 Speaker 1: oversimplified explanation, but it kind of gets the idea across. 504 00:31:54,760 --> 00:31:57,440 Speaker 1: Artificial neural networks are often used in the field of 505 00:31:57,520 --> 00:32:01,680 Speaker 1: machine learning, in which researchers train computer system to produce 506 00:32:01,800 --> 00:32:06,080 Speaker 1: specific results given specific input. Now, that could be as 507 00:32:06,120 --> 00:32:11,240 Speaker 1: simple as indicating which of a million different photographs are 508 00:32:11,320 --> 00:32:14,040 Speaker 1: the ones that happen to have cats in them versus 509 00:32:14,160 --> 00:32:16,440 Speaker 1: ones that don't have cats in them, or it could 510 00:32:16,480 --> 00:32:21,200 Speaker 1: be something far more complicated, like learning which environmental factors 511 00:32:21,400 --> 00:32:24,880 Speaker 1: impact the development of weather systems so that you can 512 00:32:24,960 --> 00:32:29,440 Speaker 1: have a more accurate weather forecast. And NPU is tuned 513 00:32:29,560 --> 00:32:33,120 Speaker 1: to work in this discipline, and often it could produce 514 00:32:33,240 --> 00:32:37,200 Speaker 1: much better results than a GPU. Both an NPU and 515 00:32:37,320 --> 00:32:40,520 Speaker 1: a GPU tend to be made with parallel processing in mind, 516 00:32:40,640 --> 00:32:45,640 Speaker 1: and NPUs are typically incorporated onto integrated circuits that also 517 00:32:45,680 --> 00:32:50,160 Speaker 1: have a CPU. They don't necessarily replace a CPU, they 518 00:32:50,160 --> 00:32:53,960 Speaker 1: are in addition to one. So let's wrestle this all 519 00:32:54,000 --> 00:32:58,680 Speaker 1: back to artificial intelligence. When you hear the phrase AI chip, 520 00:32:59,080 --> 00:33:02,600 Speaker 1: chances are the chip question is one of four types. 521 00:33:03,080 --> 00:33:09,000 Speaker 1: It's an FPGA, an ASIC, an NPU, or a GPU. 522 00:33:09,480 --> 00:33:12,720 Speaker 1: Now you can have AI enabled CPUs that don't have 523 00:33:13,360 --> 00:33:16,800 Speaker 1: these other components. But the problem with CPUs is that 524 00:33:16,920 --> 00:33:20,400 Speaker 1: due to their unspecialized design, they have limited usefulness when 525 00:33:20,440 --> 00:33:24,320 Speaker 1: it comes to AI applications, particularly as the AI field 526 00:33:24,360 --> 00:33:29,560 Speaker 1: becomes more sophisticated and has greater data processing needs. It's 527 00:33:29,640 --> 00:33:32,680 Speaker 1: kind of like giving a really good third year math 528 00:33:32,760 --> 00:33:36,480 Speaker 1: student a challenging quiz meant for fifth year students. Our 529 00:33:36,520 --> 00:33:39,120 Speaker 1: little test subject might do a decent job at the 530 00:33:39,160 --> 00:33:41,320 Speaker 1: end of the day, but it will likely take them 531 00:33:41,360 --> 00:33:44,280 Speaker 1: longer and cause more exertion than it would for someone 532 00:33:44,360 --> 00:33:48,000 Speaker 1: who is more attuned to the task. So with CPUs, 533 00:33:48,360 --> 00:33:51,760 Speaker 1: that means that you have to have longer processing times 534 00:33:51,800 --> 00:33:55,400 Speaker 1: and you have to use more energy in order to 535 00:33:55,440 --> 00:33:58,880 Speaker 1: be able to complete the task, and that means also 536 00:33:58,960 --> 00:34:03,640 Speaker 1: generating more heat. It's less efficient, it's less money efficient 537 00:34:03,680 --> 00:34:07,720 Speaker 1: as well, not just power efficient, but financially efficient. So 538 00:34:07,840 --> 00:34:10,840 Speaker 1: two of the components you find on these integrated circuits 539 00:34:10,960 --> 00:34:14,920 Speaker 1: are logic gates and we could just call them transistors 540 00:34:14,960 --> 00:34:18,960 Speaker 1: for simplicity, and then memory. So while a CPU depends 541 00:34:19,000 --> 00:34:21,040 Speaker 1: on both of these quite a bit in order to 542 00:34:21,040 --> 00:34:26,480 Speaker 1: do its job, specialized chips like ASICs AASEYS can be 543 00:34:26,520 --> 00:34:30,640 Speaker 1: made to emphasize the logic components more than the memory components, 544 00:34:30,680 --> 00:34:34,160 Speaker 1: and they can be packed with more transistors with less 545 00:34:34,200 --> 00:34:37,759 Speaker 1: space reserved for memory. That's typically what AI needs needs 546 00:34:37,800 --> 00:34:41,960 Speaker 1: access to large capacity for data processing, so the goal 547 00:34:42,040 --> 00:34:45,160 Speaker 1: is to allow for more data processing per unit of 548 00:34:45,400 --> 00:34:48,640 Speaker 1: energy than you would get out of a typical microchip. 549 00:34:49,320 --> 00:34:53,960 Speaker 1: AI is a power hungry technology. I mean that literally. 550 00:34:54,880 --> 00:34:57,960 Speaker 1: Maybe one day AI will be power hungry in the 551 00:34:58,160 --> 00:35:03,040 Speaker 1: figurative sense, like in like the super villain sense. Maybe 552 00:35:03,040 --> 00:35:05,319 Speaker 1: that will happen one day, but right now, it's just 553 00:35:05,400 --> 00:35:08,680 Speaker 1: it needs a lot of juice. So making the processing 554 00:35:08,719 --> 00:35:12,840 Speaker 1: as efficient as possible is absolutely vital, Otherwise the costs 555 00:35:12,840 --> 00:35:16,680 Speaker 1: of operations spiral out of control. Your energy needs as 556 00:35:16,680 --> 00:35:19,239 Speaker 1: well as things like cooling needs and everything else that 557 00:35:19,280 --> 00:35:22,600 Speaker 1: goes along with using a bucket load of power would 558 00:35:22,600 --> 00:35:25,440 Speaker 1: make it harder for you to cover costs. This is 559 00:35:25,480 --> 00:35:28,640 Speaker 1: part of the reason why you'll hear about companies spending 560 00:35:28,719 --> 00:35:32,560 Speaker 1: billions of dollars on AI. It's not just that they 561 00:35:32,560 --> 00:35:35,080 Speaker 1: have to spend that money for the research and development 562 00:35:35,080 --> 00:35:37,759 Speaker 1: of AI, although that takes up a big part of it. 563 00:35:37,760 --> 00:35:41,920 Speaker 1: It's that actually operating these data centers that are running 564 00:35:41,960 --> 00:35:45,440 Speaker 1: these specialized machines takes a lot of energy, and so 565 00:35:45,719 --> 00:35:50,480 Speaker 1: the cost of operation is in the billions of dollars. Now, 566 00:35:50,760 --> 00:35:54,360 Speaker 1: these AI chips typically can handle parallel processing tasks in 567 00:35:54,400 --> 00:35:57,360 Speaker 1: a much greater capacity than even your most powerful multi 568 00:35:57,360 --> 00:36:00,959 Speaker 1: thread into multi core CPUs can. Which type of chip 569 00:36:01,000 --> 00:36:05,440 Speaker 1: you use often depends upon the application you want, so, 570 00:36:05,600 --> 00:36:11,160 Speaker 1: for example, Google's tensor processing Unit is an ASK chip. 571 00:36:11,520 --> 00:36:14,320 Speaker 1: Google has spent a lot of time and money developing 572 00:36:14,360 --> 00:36:18,000 Speaker 1: these processors and fine tuning them to handle intense data 573 00:36:18,080 --> 00:36:23,040 Speaker 1: processing at incredible speed for the purposes of machine learning applications. Primarily, 574 00:36:23,400 --> 00:36:26,000 Speaker 1: a lot of AI companies will use off the shelf 575 00:36:26,040 --> 00:36:29,719 Speaker 1: GPUs and they will wire them together in order to 576 00:36:29,800 --> 00:36:33,560 Speaker 1: train AI models, which has led to Nvidia, which for 577 00:36:33,719 --> 00:36:36,600 Speaker 1: years was thought of as just a graphics processing unit 578 00:36:36,719 --> 00:36:41,120 Speaker 1: design company, to now become a leading AI chip company. 579 00:36:41,680 --> 00:36:45,440 Speaker 1: The boom in AI development has catapulted Nvidia to become 580 00:36:45,480 --> 00:36:49,600 Speaker 1: a three trillion dollar company in recent years, so it 581 00:36:49,640 --> 00:36:52,399 Speaker 1: has joined the likes of Microsoft and Apple. That's not 582 00:36:52,400 --> 00:36:55,000 Speaker 1: to say Nvidia was always like an underdog or anything. 583 00:36:55,040 --> 00:36:58,560 Speaker 1: It was always a company that was doing pretty darn well, 584 00:36:59,040 --> 00:37:03,600 Speaker 1: but in recent years it has entered the stratospheric level evaluation. 585 00:37:04,400 --> 00:37:07,839 Speaker 1: It was not a trillion dollar company that long ago. 586 00:37:08,000 --> 00:37:12,080 Speaker 1: When we talk about consumer products, CPUs and NPUs are 587 00:37:12,120 --> 00:37:17,480 Speaker 1: typically what will handle AI needs because they are the 588 00:37:17,480 --> 00:37:21,640 Speaker 1: more cost efficient approach. Intel has developed NPUs under the 589 00:37:21,640 --> 00:37:25,040 Speaker 1: code name metior Lake. Actually, to be more precise, the 590 00:37:25,120 --> 00:37:28,880 Speaker 1: metior Lake chips include CPU cores. They also include a 591 00:37:28,920 --> 00:37:33,040 Speaker 1: small GPU portion as well as the NPU unit, all 592 00:37:33,200 --> 00:37:36,520 Speaker 1: on this same integrated circuit. And the idea is that 593 00:37:36,560 --> 00:37:39,399 Speaker 1: these chips will be incorporated into machines that can run 594 00:37:39,440 --> 00:37:43,480 Speaker 1: AI workloads locally. So let's say you've got a company 595 00:37:43,760 --> 00:37:46,560 Speaker 1: and that company wants to host a language model, but 596 00:37:46,640 --> 00:37:49,080 Speaker 1: it wants it locally. It doesn't want to be tapping 597 00:37:49,120 --> 00:37:52,680 Speaker 1: into a cloud based language model, they want to run 598 00:37:52,680 --> 00:37:56,480 Speaker 1: it on premises, while they might use computers with meteor 599 00:37:56,560 --> 00:37:59,319 Speaker 1: Lake chips in them in order to do that processing, 600 00:37:59,400 --> 00:38:01,600 Speaker 1: which would be more cost effective than building out a 601 00:38:01,640 --> 00:38:06,200 Speaker 1: whole AI data center just to service this specific company. Okay, 602 00:38:06,520 --> 00:38:09,759 Speaker 1: so when people talk about AI chips, they don't mean 603 00:38:09,840 --> 00:38:14,280 Speaker 1: that somehow the chips are imbued with artificial intelligence. Instead, 604 00:38:14,320 --> 00:38:18,680 Speaker 1: these chips are optimized to run AI applications, and those 605 00:38:18,719 --> 00:38:23,000 Speaker 1: applications run the entire gamut of AI. There are AI 606 00:38:23,080 --> 00:38:26,400 Speaker 1: chips used in robotics, There are AI chips used in 607 00:38:26,480 --> 00:38:31,280 Speaker 1: autonomous cars. There are AI chips for large language models. 608 00:38:31,600 --> 00:38:35,360 Speaker 1: Smaller chips and NPUs can be incorporated into smart devices, 609 00:38:35,400 --> 00:38:39,040 Speaker 1: which allow some AI processing to happen at the device 610 00:38:39,160 --> 00:38:43,200 Speaker 1: level rather than remotely through a network connection. That's really 611 00:38:43,239 --> 00:38:47,080 Speaker 1: important for speeding up those processes and to eliminate latency, 612 00:38:47,480 --> 00:38:51,799 Speaker 1: because for some implementations speed might not be that big 613 00:38:51,840 --> 00:38:55,720 Speaker 1: a deal, but for others, like the autonomous cars I mentioned, 614 00:38:56,000 --> 00:38:59,680 Speaker 1: being able to process information and produce results is critical 615 00:38:59,760 --> 00:39:03,319 Speaker 1: to operate the technology safely. You cannot have latency in 616 00:39:03,360 --> 00:39:07,080 Speaker 1: those systems or disaster can occur. You wouldn't want an 617 00:39:07,080 --> 00:39:10,440 Speaker 1: autonomous car that constantly has to beam information up to 618 00:39:10,480 --> 00:39:13,720 Speaker 1: the cloud and wait for a response, because real world 619 00:39:14,040 --> 00:39:19,120 Speaker 1: driving conditions are constantly changing. They are dynamic, and they 620 00:39:19,239 --> 00:39:22,600 Speaker 1: change at a very fast rate. Depending on how quickly 621 00:39:22,600 --> 00:39:25,600 Speaker 1: you're driving, it could be an incredibly fast rate. So 622 00:39:26,000 --> 00:39:30,920 Speaker 1: any latency would lead to catastrophic outcomes. So AI chips 623 00:39:30,960 --> 00:39:35,759 Speaker 1: are important components in what you might call EDGEAI. This 624 00:39:35,840 --> 00:39:38,279 Speaker 1: not only cuts down on processing time, but it can 625 00:39:38,360 --> 00:39:42,480 Speaker 1: also help things remain more secure. Right, you're not beaming 626 00:39:42,600 --> 00:39:45,680 Speaker 1: data to a different location all the time, you're processing 627 00:39:45,680 --> 00:39:51,320 Speaker 1: it locally. That makes it less susceptible to being hacked. 628 00:39:51,800 --> 00:39:56,160 Speaker 1: Not immune, but it's less susceptible. There's fewer links in 629 00:39:56,200 --> 00:39:59,560 Speaker 1: the chain, you could say. So now we have our 630 00:39:59,600 --> 00:40:02,560 Speaker 1: overview of what AI chips are all about. And I 631 00:40:02,600 --> 00:40:05,640 Speaker 1: think it's good to remember that a processor's utility depends 632 00:40:05,800 --> 00:40:09,040 Speaker 1: entirely upon what you plan to actually use it for. 633 00:40:09,400 --> 00:40:12,200 Speaker 1: If you're doing standard computing stuff like you're working with 634 00:40:12,280 --> 00:40:16,440 Speaker 1: documents or playing games or browsing the web, and AI 635 00:40:16,560 --> 00:40:19,360 Speaker 1: chip isn't really going to mean much to you at all. 636 00:40:19,880 --> 00:40:23,400 Speaker 1: AI chips tend to be really geared toward parallel processing. 637 00:40:23,719 --> 00:40:27,560 Speaker 1: So it's possible that a computer with a good AI 638 00:40:27,680 --> 00:40:30,960 Speaker 1: chip could be useful as a gaming rig. But honestly, 639 00:40:31,040 --> 00:40:33,840 Speaker 1: I think, at least for now, going with a good 640 00:40:33,920 --> 00:40:38,720 Speaker 1: GPU and a decent CPU matters more for gamers, And 641 00:40:39,160 --> 00:40:42,200 Speaker 1: like I said, some cases, you might not need a 642 00:40:42,239 --> 00:40:45,520 Speaker 1: really good GPU. You could have a decent GPU and 643 00:40:45,760 --> 00:40:48,080 Speaker 1: a really good CPU. It all kind of depends on 644 00:40:48,120 --> 00:40:50,399 Speaker 1: the types of games you want to play. I think 645 00:40:50,400 --> 00:40:52,880 Speaker 1: it's important for regular old folks like me and at 646 00:40:52,960 --> 00:40:55,319 Speaker 1: least some of y'all out there, to know about this 647 00:40:55,360 --> 00:40:59,040 Speaker 1: stuff so that when we're shopping around for our next device, 648 00:40:59,600 --> 00:41:02,799 Speaker 1: we have an understanding of the terminology. Right, we know 649 00:41:02,840 --> 00:41:05,640 Speaker 1: what an AI chip is and what it's supposed to do, 650 00:41:05,760 --> 00:41:08,640 Speaker 1: and whether or not it matches what we need. We 651 00:41:08,680 --> 00:41:12,200 Speaker 1: aren't just pooled by marketing terms. You know. It doesn't 652 00:41:12,520 --> 00:41:15,400 Speaker 1: mean that an AI chip labels slapped on something is 653 00:41:15,400 --> 00:41:18,080 Speaker 1: going to mean that that's the best thing for us. 654 00:41:18,600 --> 00:41:22,400 Speaker 1: So having this understanding is important. Being an informed consumer 655 00:41:22,560 --> 00:41:25,880 Speaker 1: is important. It means you're going to get the best 656 00:41:26,680 --> 00:41:30,920 Speaker 1: out of your money that meets your needs. Right, we 657 00:41:31,000 --> 00:41:33,040 Speaker 1: only have so much money. We should make sure that 658 00:41:33,080 --> 00:41:35,680 Speaker 1: when we're spending it. We're doing it on stuff that 659 00:41:35,800 --> 00:41:38,960 Speaker 1: actually solves the problems we have, as opposed to just 660 00:41:39,320 --> 00:41:42,239 Speaker 1: stuff that's shiny and new. I say this because a 661 00:41:42,280 --> 00:41:45,400 Speaker 1: lot of tech enthusiasts tend to fall into the trap 662 00:41:45,440 --> 00:41:48,640 Speaker 1: of I want the new thing because the new thing 663 00:41:48,840 --> 00:41:52,000 Speaker 1: is somehow better than the old thing. That's not always 664 00:41:52,040 --> 00:41:55,120 Speaker 1: the case. It often is in tech, but it's not 665 00:41:55,200 --> 00:41:58,680 Speaker 1: always the case, and it certainly doesn't always justify spending 666 00:41:58,680 --> 00:42:00,680 Speaker 1: the amount of money it takes to be part of 667 00:42:00,719 --> 00:42:03,560 Speaker 1: that bleeding edge. It is important that we have a 668 00:42:03,600 --> 00:42:06,839 Speaker 1: bleeding edge, but it's not important that we're all in it. 669 00:42:07,480 --> 00:42:09,160 Speaker 1: We can hang back a bit if we need to. 670 00:42:09,600 --> 00:42:12,919 Speaker 1: So I just wanted to take this chance to kind 671 00:42:12,920 --> 00:42:16,440 Speaker 1: of break down this AI chip terminology and what it 672 00:42:16,520 --> 00:42:20,239 Speaker 1: actually means, because goodness knows, Like when I started first 673 00:42:20,280 --> 00:42:24,440 Speaker 1: seeing the terminology myself, I was confused. I was thinking, 674 00:42:24,480 --> 00:42:28,360 Speaker 1: what makes an AI chip and aichip? And does it 675 00:42:28,440 --> 00:42:31,640 Speaker 1: have some sort of AI capability built into it? Because 676 00:42:31,680 --> 00:42:35,120 Speaker 1: how would that work? And obviously I was overthinking it. 677 00:42:35,360 --> 00:42:38,200 Speaker 1: So hopefully this was useful for y'all, and I hope 678 00:42:38,200 --> 00:42:40,920 Speaker 1: you're all doing well, and I'll talk to you again 679 00:42:41,640 --> 00:42:51,799 Speaker 1: really soon. Tech Stuff is an iHeartRadio production. For more 680 00:42:51,840 --> 00:42:56,600 Speaker 1: podcasts from iHeartRadio, visit the iHeartRadio app, Apple Podcasts, or 681 00:42:56,600 --> 00:42:58,560 Speaker 1: wherever you listen to your favorite shows.