1 00:00:00,000 --> 00:00:01,920 All right, so get this. 2 00:00:01,920 --> 00:00:06,720 Those amazing graphics you see in games like Cyberpunk 2077, 3 00:00:06,720 --> 00:00:09,360 that's your graphics card churning out 4 00:00:09,360 --> 00:00:13,000 an insane number of calculations every second. 5 00:00:13,000 --> 00:00:13,640 How many? 6 00:00:13,640 --> 00:00:17,360 Like a mind-boggling 36 trillion. 7 00:00:17,360 --> 00:00:19,760 Wow, 36 trillion every second. 8 00:00:19,760 --> 00:00:20,720 That's incredible. 9 00:00:20,720 --> 00:00:25,920 To put that into perspective, Mario 64 on the Nintendo 64 10 00:00:25,920 --> 00:00:29,680 only needed about 100 million calculations per second. 11 00:00:29,680 --> 00:00:30,200 Oh, wow. 12 00:00:30,200 --> 00:00:36,040 And even Minecraft back in 2011 was only around 100 billion. 13 00:00:36,040 --> 00:00:39,000 That's a huge jump in processing power, right? 14 00:00:39,000 --> 00:00:40,440 Absolutely, a massive leap. 15 00:00:40,440 --> 00:00:41,960 Yeah, so in this deep dive, we're 16 00:00:41,960 --> 00:00:44,280 going to be using the sources that you provided 17 00:00:44,280 --> 00:00:47,080 to really unpack how graphics cards evolved 18 00:00:47,080 --> 00:00:50,480 into these computational behemoths. 19 00:00:50,480 --> 00:00:52,840 It really is a fascinating evolution. 20 00:00:52,840 --> 00:00:55,480 We'll be exploring everything from the physical components 21 00:00:55,480 --> 00:00:58,600 to the really mind-bending architecture that's 22 00:00:58,600 --> 00:01:00,000 behind all those calculations. 23 00:01:00,000 --> 00:01:02,680 And you actually provided some really interesting research 24 00:01:02,680 --> 00:01:06,040 on the fundamental differences between a CPU, 25 00:01:06,040 --> 00:01:08,520 the brains of your computer, and a GTU. 26 00:01:08,520 --> 00:01:11,800 Right, yeah, it's a common point of confusion. 27 00:01:11,800 --> 00:01:14,080 They both handle calculations, obviously. 28 00:01:14,080 --> 00:01:16,320 But they do it in very different ways. 29 00:01:16,320 --> 00:01:22,560 A CPU is like a highly skilled artisan, very fast, 30 00:01:22,560 --> 00:01:27,760 very versatile, but can only work on one or a few tasks 31 00:01:27,760 --> 00:01:31,680 at a time, whereas a GPU is like a massive factory 32 00:01:31,680 --> 00:01:36,680 with thousands of workers all doing relatively simple tasks, 33 00:01:36,680 --> 00:01:38,840 but simultaneously. 34 00:01:38,840 --> 00:01:43,280 So when you need to render millions of pixels in a game, 35 00:01:43,280 --> 00:01:48,320 the GPU's brute force approach of parallel processing 36 00:01:48,320 --> 00:01:49,080 wins out. 37 00:01:49,080 --> 00:01:52,240 Exactly, it's all about tackling what we call embarrassingly 38 00:01:52,240 --> 00:01:53,920 parallel problems. 39 00:01:53,920 --> 00:01:57,680 Things like processing image data or simulating physics. 40 00:01:57,680 --> 00:02:00,160 Where you can break the task down into millions 41 00:02:00,160 --> 00:02:01,600 of independent calculations. 42 00:02:01,600 --> 00:02:05,480 OK, so let's peek inside one of these GPU factories. 43 00:02:05,480 --> 00:02:07,400 What are we looking at here? 44 00:02:07,400 --> 00:02:11,000 Well, at the heart of it all is the graphics processing 45 00:02:11,000 --> 00:02:12,600 unit or the GPU. 46 00:02:12,600 --> 00:02:13,480 The GPU, yeah. 47 00:02:13,480 --> 00:02:16,880 And we often refer to the whole graphics card as a GPU. 48 00:02:16,880 --> 00:02:20,600 But it's actually just one component on a circuit board 49 00:02:20,600 --> 00:02:24,080 with its own power delivery system, cooling, 50 00:02:24,080 --> 00:02:25,160 and of course, memory. 51 00:02:25,160 --> 00:02:25,520 I see. 52 00:02:25,520 --> 00:02:29,640 And in the sources you shared, some schematics and diagrams, 53 00:02:29,640 --> 00:02:33,680 one thing that really stands out is this GA-102 chip, 54 00:02:33,680 --> 00:02:35,240 the brain of the graphics card. 55 00:02:35,240 --> 00:02:41,960 Yes, this chip alone can contain over 28 billion transistors, 56 00:02:41,960 --> 00:02:46,240 which is just an astounding feat of engineering. 57 00:02:46,240 --> 00:02:48,760 Wow, that's more than. 58 00:02:48,760 --> 00:02:50,720 To give you an idea of scale, that's 59 00:02:50,720 --> 00:02:52,680 like fitting the population of the entire planet 60 00:02:52,680 --> 00:02:53,760 onto your fingernail. 61 00:02:53,760 --> 00:02:54,240 Oh, wow. 62 00:02:54,240 --> 00:02:56,800 And all those transistors are organized 63 00:02:56,800 --> 00:02:58,600 in a hierarchical structure, right? 64 00:02:58,600 --> 00:02:59,640 Yes, that's right. 65 00:02:59,640 --> 00:03:01,880 You have these larger units called graphics processing 66 00:03:01,880 --> 00:03:04,120 clusters, which are further divided 67 00:03:04,120 --> 00:03:07,360 into streaming multiprocessors. 68 00:03:07,360 --> 00:03:10,320 And within each of those, you have the individual cores 69 00:03:10,320 --> 00:03:12,200 that are doing the actual calculations. 70 00:03:12,200 --> 00:03:15,200 So remind me again, what are the different types of cores? 71 00:03:15,200 --> 00:03:17,000 And what do they actually do? 72 00:03:17,000 --> 00:03:21,160 So the workhorses for most gaming tasks 73 00:03:21,160 --> 00:03:23,680 are the CUA cores. 74 00:03:23,680 --> 00:03:26,480 Think of them as the general purpose workers 75 00:03:26,480 --> 00:03:28,040 in our factory analogy. 76 00:03:28,040 --> 00:03:31,280 Then you have tensor cores, which are specialized 77 00:03:31,280 --> 00:03:33,600 for matrix math. 78 00:03:33,600 --> 00:03:36,840 They're particularly crucial for AI applications 79 00:03:36,840 --> 00:03:40,440 because of their ability to process huge matrices, which 80 00:03:40,440 --> 00:03:43,440 are the building blocks of many AI algorithms. 81 00:03:43,440 --> 00:03:45,120 And then there are ray tracing cores, 82 00:03:45,120 --> 00:03:47,560 which I know are all the rage in gaming these days. 83 00:03:47,560 --> 00:03:51,000 Yes, ray tracing is a computationally intensive 84 00:03:51,000 --> 00:03:54,480 technique that simulates the path of light 85 00:03:54,480 --> 00:03:57,960 to create incredibly realistic lighting and reflections 86 00:03:57,960 --> 00:03:59,440 in games. 87 00:03:59,440 --> 00:04:02,000 It's been around for a while in other industries like film, 88 00:04:02,000 --> 00:04:06,400 but only recently have GPUs become powerful enough 89 00:04:06,400 --> 00:04:09,360 to handle real-time retracing in games. 90 00:04:09,360 --> 00:04:11,160 Now, here's something I found really interesting 91 00:04:11,160 --> 00:04:13,160 in the materials you shared. 92 00:04:13,160 --> 00:04:15,440 The concept of binning. 93 00:04:15,440 --> 00:04:19,040 It seems like the same GA-102 chip 94 00:04:19,040 --> 00:04:21,600 can end up in a variety of different graphics cards, 95 00:04:21,600 --> 00:04:24,880 like the 3080, the 3090, and their tie variants. 96 00:04:24,880 --> 00:04:27,640 The binning is a fascinating process. 97 00:04:27,640 --> 00:04:30,880 Essentially, not every chip manufactured is perfect. 98 00:04:30,880 --> 00:04:33,760 Some might have tiny flaws or defects. 99 00:04:33,760 --> 00:04:36,320 The chips with the fewest flaws are 100 00:04:36,320 --> 00:04:40,800 sorted into the higher-end cards, like the 3090 Ti, which 101 00:04:40,800 --> 00:04:46,240 boasts a full complement of CUDA cores and higher clock speeds. 102 00:04:46,240 --> 00:04:49,480 Cards like the 3080 might have some cores deactivated 103 00:04:49,480 --> 00:04:51,680 due to these minor flaws. 104 00:04:51,680 --> 00:04:53,560 So it's a way to maximize the yield 105 00:04:53,560 --> 00:04:57,360 from the manufacturing process and create a range of products 106 00:04:57,360 --> 00:04:58,840 at different price points. 107 00:04:58,840 --> 00:05:01,920 But how do they actually test and sort 108 00:05:01,920 --> 00:05:03,840 these billions of transistors? 109 00:05:03,840 --> 00:05:06,120 It's a highly automated process that 110 00:05:06,120 --> 00:05:08,720 involves running a series of tests 111 00:05:08,720 --> 00:05:12,360 to identify any defects or performance limitations. 112 00:05:12,360 --> 00:05:16,040 The chips are then categorized based on their capabilities. 113 00:05:16,040 --> 00:05:20,440 And the best ones go to the top tier cards. 114 00:05:20,440 --> 00:05:20,920 OK. 115 00:05:20,920 --> 00:05:23,600 So we've got these different cores doing their thing. 116 00:05:23,600 --> 00:05:26,600 But they need a constant stream of data to work on, right? 117 00:05:26,600 --> 00:05:27,600 Absolutely, yes. 118 00:05:27,600 --> 00:05:29,240 That's where graphics memory comes in? 119 00:05:29,240 --> 00:05:30,280 That's where it comes in. 120 00:05:30,280 --> 00:05:35,160 GDDR6XSDRAM is the latest generation of graphics memory. 121 00:05:35,160 --> 00:05:40,800 And it plays a crucial role in seeding the GPU with the data 122 00:05:40,800 --> 00:05:42,000 it needs. 123 00:05:42,000 --> 00:05:44,400 Think of it as a high-speed conveyor belt, 124 00:05:44,400 --> 00:05:47,080 delivering raw materials to our factory workers. 125 00:05:47,080 --> 00:05:49,800 And it's not just about the amount of memory, right? 126 00:05:49,800 --> 00:05:53,560 But also the speed at which data can be transferred. 127 00:05:53,560 --> 00:05:54,160 Yes. 128 00:05:54,160 --> 00:05:56,080 Bandwidth is the term that keeps popping up. 129 00:05:56,080 --> 00:05:57,520 Bandwidth is crucial. 130 00:05:57,520 --> 00:06:01,400 It's a measure of how much data can be transferred per second. 131 00:06:01,400 --> 00:06:05,680 GDDR6X uses some clever techniques, 132 00:06:05,680 --> 00:06:09,800 like PAM3 and PAM4 encoding, to cram more data 133 00:06:09,800 --> 00:06:11,840 through the same physical connections. 134 00:06:11,840 --> 00:06:14,600 Instead of just using zeros and one, 135 00:06:14,600 --> 00:06:18,920 they use multiple voltage levels to represent more information. 136 00:06:18,920 --> 00:06:24,320 So PAM3 is the newer and more efficient encoding scheme, right? 137 00:06:24,320 --> 00:06:24,680 That's right. 138 00:06:24,680 --> 00:06:28,080 PAM3 was adopted for its better signal integrity 139 00:06:28,080 --> 00:06:30,680 and efficiency, even though it technically 140 00:06:30,680 --> 00:06:33,800 uses fewer voltage levels than PAM4. 141 00:06:33,800 --> 00:06:37,840 It's a trade-off between speed and reliability. 142 00:06:37,840 --> 00:06:40,640 There's a constant push and pull between these factors 143 00:06:40,640 --> 00:06:45,600 as engineers strive to squeeze every bit of performance 144 00:06:45,600 --> 00:06:46,680 out of these systems. 145 00:06:46,680 --> 00:06:49,160 OK, so we've covered the basics of the GPU. 146 00:06:49,160 --> 00:06:49,520 Yes. 147 00:06:49,520 --> 00:06:52,480 It's cores and how data flows in and out. 148 00:06:52,480 --> 00:06:54,600 But now we need to dive into the real magic, 149 00:06:54,600 --> 00:06:59,160 like the architecture that allows these billions of transistors 150 00:06:59,160 --> 00:07:01,080 to work together in parallel. 151 00:07:01,080 --> 00:07:03,080 That's where things get really interesting. 152 00:07:03,080 --> 00:07:04,080 Yeah. 153 00:07:04,080 --> 00:07:07,280 We'll be exploring concepts like SIMD and SIMT, 154 00:07:07,280 --> 00:07:11,480 which are fundamental to understanding how GPUs achieve 155 00:07:11,480 --> 00:07:12,920 their incredible performance. 156 00:07:12,920 --> 00:07:14,000 Let's take a break there. 157 00:07:14,000 --> 00:07:14,360 OK. 158 00:07:14,360 --> 00:07:16,840 And come back to unravel those acronyms. 159 00:07:16,840 --> 00:07:17,360 That's good. 160 00:07:17,360 --> 00:07:18,120 In part two. 161 00:07:18,120 --> 00:07:18,760 All right, all right. 162 00:07:18,760 --> 00:07:21,240 This stuff is truly mind-blowing. 163 00:07:21,240 --> 00:07:21,840 It is. 164 00:07:21,840 --> 00:07:22,040 It is. 165 00:07:22,040 --> 00:07:22,640 Welcome back. 166 00:07:22,640 --> 00:07:25,400 So last time, we were exploring the physical heart 167 00:07:25,400 --> 00:07:26,560 of a graphics card. 168 00:07:26,560 --> 00:07:27,800 Right, all those components. 169 00:07:27,800 --> 00:07:30,320 Yeah, and now we're going even deeper. 170 00:07:30,320 --> 00:07:31,080 Oh, wow. 171 00:07:31,080 --> 00:07:31,680 Deeper. 172 00:07:31,680 --> 00:07:34,840 Yeah, to the level of the instructions 173 00:07:34,840 --> 00:07:38,560 that tell all those billions of transistors what to do. 174 00:07:38,560 --> 00:07:40,240 So like the software almost. 175 00:07:40,240 --> 00:07:41,320 Yeah, kind of, yeah. 176 00:07:41,320 --> 00:07:41,880 OK. 177 00:07:41,880 --> 00:07:45,240 So are you ready to untangle SIMD and SIMD? 178 00:07:45,240 --> 00:07:46,880 I am, and I think our listeners will 179 00:07:46,880 --> 00:07:49,680 find this particularly fascinating. 180 00:07:49,680 --> 00:07:52,960 These concepts are really key to understanding 181 00:07:52,960 --> 00:07:57,280 how GPUs achieve such mind-boggling performance. 182 00:07:57,280 --> 00:07:59,160 OK, so let's start with SIMD. 183 00:07:59,160 --> 00:07:59,840 OK. 184 00:07:59,840 --> 00:08:03,520 Single instruction, multiple data, 185 00:08:03,520 --> 00:08:05,560 sounds kind of like doing a lot with a little. 186 00:08:05,560 --> 00:08:06,800 That's the essence of it, yeah. 187 00:08:06,800 --> 00:08:07,520 OK. 188 00:08:07,520 --> 00:08:09,360 Imagine you have a recipe. 189 00:08:09,360 --> 00:08:10,720 But instead of making one dish, you 190 00:08:10,720 --> 00:08:15,040 could make dozens simultaneously using the same instructions. 191 00:08:15,040 --> 00:08:16,640 SIMD works on that principle. 192 00:08:16,640 --> 00:08:19,680 So instead of processing one data point at a time, 193 00:08:19,680 --> 00:08:23,320 a SIMD instruction can operate on a whole chunk of data 194 00:08:23,320 --> 00:08:24,400 simultaneously. 195 00:08:24,400 --> 00:08:25,440 Exactly. 196 00:08:25,440 --> 00:08:28,640 Think about a simple operation, like adding five 197 00:08:28,640 --> 00:08:30,840 to every number in a list. 198 00:08:30,840 --> 00:08:34,200 A CPU would have to loop through each number individually. 199 00:08:34,200 --> 00:08:40,360 A GPU using SIMD could add five to, say, 32 numbers at once, 200 00:08:40,360 --> 00:08:41,640 depending on its architecture. 201 00:08:41,640 --> 00:08:42,840 OK, that makes sense. 202 00:08:42,840 --> 00:08:43,480 Yeah. 203 00:08:43,480 --> 00:08:46,640 But how does this translate to the complex world 204 00:08:46,640 --> 00:08:47,960 of graphics rendering? 205 00:08:47,960 --> 00:08:50,560 OK, let's go back to the example of transforming 206 00:08:50,560 --> 00:08:53,680 vertex coordinates, which defines the shape of objects 207 00:08:53,680 --> 00:08:54,920 in a game world. 208 00:08:54,920 --> 00:08:58,080 Each vertex has an x, y, and z coordinate 209 00:08:58,080 --> 00:09:00,200 to place that vertex correctly. 210 00:09:00,200 --> 00:09:02,160 You need to apply a matrix transformation. 211 00:09:02,160 --> 00:09:02,960 OK. 212 00:09:02,960 --> 00:09:06,800 With SIMD, the GPU can perform that same transformation 213 00:09:06,800 --> 00:09:08,880 on multiple vertices at once, right, 214 00:09:08,880 --> 00:09:10,760 drastically speeding up the process. 215 00:09:10,760 --> 00:09:13,200 Yeah, I'm starting to see why this is so crucial. 216 00:09:13,200 --> 00:09:13,600 Right. 217 00:09:13,600 --> 00:09:17,880 For handling millions of pixels and polygons. 218 00:09:17,880 --> 00:09:18,560 It's exactly. 219 00:09:18,560 --> 00:09:19,280 In real time. 220 00:09:19,280 --> 00:09:21,920 Yeah, it's all about leveraging parallelism. 221 00:09:21,920 --> 00:09:24,600 But SIMD has its limitations. 222 00:09:24,600 --> 00:09:25,320 OK. 223 00:09:25,320 --> 00:09:28,960 It assumes all data needs the same operation applied to it. 224 00:09:28,960 --> 00:09:29,600 OK. 225 00:09:29,600 --> 00:09:30,760 That's where SIMD comes in. 226 00:09:30,760 --> 00:09:31,320 OK. 227 00:09:31,320 --> 00:09:33,840 Single instruction, multiple threads. 228 00:09:33,840 --> 00:09:36,680 And from what I gathered in the research that you provided, 229 00:09:36,680 --> 00:09:37,180 yeah. 230 00:09:37,180 --> 00:09:39,040 SIMD seems to be more flexible. 231 00:09:39,040 --> 00:09:39,720 You nailed it. 232 00:09:39,720 --> 00:09:40,960 OK. 233 00:09:40,960 --> 00:09:43,960 SIMD allows different threads within a group. 234 00:09:43,960 --> 00:09:44,400 Right. 235 00:09:44,400 --> 00:09:47,080 To execute instructions at their own pace. 236 00:09:47,080 --> 00:09:47,720 OK. 237 00:09:47,720 --> 00:09:49,400 Even if they're all technically working 238 00:09:49,400 --> 00:09:50,640 from the same instruction set. 239 00:09:50,640 --> 00:09:51,440 Right. 240 00:09:51,440 --> 00:09:56,000 This is essential for dealing with real world scenarios. 241 00:09:56,000 --> 00:09:56,360 Right. 242 00:09:56,360 --> 00:09:58,800 Where not all data needs the same treatment. 243 00:09:58,800 --> 00:10:00,960 So if some threads need to take a detour, 244 00:10:00,960 --> 00:10:02,360 like in their calculations. 245 00:10:02,360 --> 00:10:02,920 Right. 246 00:10:02,920 --> 00:10:04,600 They don't hold up the entire group. 247 00:10:04,600 --> 00:10:05,360 Precisely. 248 00:10:05,360 --> 00:10:05,880 OK. 249 00:10:05,880 --> 00:10:08,960 Think about rendering a complex scene with lots of objects. 250 00:10:08,960 --> 00:10:09,480 OK. 251 00:10:09,480 --> 00:10:11,160 Some might be visible to the player. 252 00:10:11,160 --> 00:10:11,640 Right. 253 00:10:11,640 --> 00:10:13,680 While others are completely obscured. 254 00:10:13,680 --> 00:10:14,080 Right. 255 00:10:14,080 --> 00:10:14,880 Behind something. 256 00:10:14,880 --> 00:10:15,400 Exactly. 257 00:10:15,400 --> 00:10:19,440 With SIMD, the GPU can intelligently allocate resources. 258 00:10:19,440 --> 00:10:19,880 Right. 259 00:10:19,880 --> 00:10:22,560 So that threads working on hidden objects 260 00:10:22,560 --> 00:10:24,360 can essentially skip. 261 00:10:24,360 --> 00:10:25,160 Oh, wow. 262 00:10:25,160 --> 00:10:26,400 Unnecessary calculations. 263 00:10:26,400 --> 00:10:27,200 I'm so smart. 264 00:10:27,200 --> 00:10:29,000 Making the whole process more efficient. 265 00:10:29,000 --> 00:10:31,000 It's like having a smart manager. 266 00:10:31,000 --> 00:10:31,400 Yes. 267 00:10:31,400 --> 00:10:33,840 Like directing traffic within the GPU. 268 00:10:33,840 --> 00:10:34,400 Right. 269 00:10:34,400 --> 00:10:36,480 Making sure things run smoothly. 270 00:10:36,480 --> 00:10:36,840 Yeah. 271 00:10:36,840 --> 00:10:38,120 Keeping everything optimized. 272 00:10:38,120 --> 00:10:38,400 Yeah. 273 00:10:38,400 --> 00:10:41,360 Even with lots of complex decisions happening. 274 00:10:41,360 --> 00:10:42,400 That's a great way to put it. 275 00:10:42,400 --> 00:10:45,440 But why wasn't SIMD always the standard? 276 00:10:45,440 --> 00:10:46,920 That's a great question. 277 00:10:46,920 --> 00:10:48,560 Like if it's more flexible. 278 00:10:48,560 --> 00:10:49,000 Right. 279 00:10:49,000 --> 00:10:50,560 Why even bother with SIMD? 280 00:10:50,560 --> 00:10:52,600 So early GPUs were simpler. 281 00:10:52,600 --> 00:10:53,160 OK. 282 00:10:53,160 --> 00:10:55,600 And relied on SIMD because it was easier 283 00:10:55,600 --> 00:10:57,080 to implement in hardware. 284 00:10:57,080 --> 00:10:57,520 Oh. 285 00:10:57,520 --> 00:10:59,280 Keeping all threads in lockstep. 286 00:10:59,280 --> 00:10:59,760 Right. 287 00:10:59,760 --> 00:11:01,360 Simplified control and management. 288 00:11:01,360 --> 00:11:01,920 Gotcha. 289 00:11:01,920 --> 00:11:04,760 But as games and graphics became more complex. 290 00:11:04,760 --> 00:11:05,720 Yeah, of course. 291 00:11:05,720 --> 00:11:08,920 The need for flexibility led to the development. 292 00:11:08,920 --> 00:11:09,320 Right. 293 00:11:09,320 --> 00:11:10,760 And adoption of SIMD. 294 00:11:10,760 --> 00:11:12,960 So SIMD is like the natural evolution. 295 00:11:12,960 --> 00:11:13,520 It is, yeah. 296 00:11:13,520 --> 00:11:14,080 Of SIMD. 297 00:11:14,080 --> 00:11:14,360 Yeah. 298 00:11:14,360 --> 00:11:15,400 You can think of it that way. 299 00:11:15,400 --> 00:11:19,200 Offering more control and efficiency for modern workloads. 300 00:11:19,200 --> 00:11:19,680 Yeah. 301 00:11:19,680 --> 00:11:21,520 For those more demanding tasks. 302 00:11:21,520 --> 00:11:24,920 It's amazing how these seemingly small changes 303 00:11:24,920 --> 00:11:26,240 in architecture. 304 00:11:26,240 --> 00:11:27,120 I know. 305 00:11:27,120 --> 00:11:29,800 Have such a huge impact on performance. 306 00:11:29,800 --> 00:11:33,600 It really is a testament to the constant innovation 307 00:11:33,600 --> 00:11:35,520 happening in the field of computer graphics. 308 00:11:35,520 --> 00:11:36,920 And speaking of innovation. 309 00:11:36,920 --> 00:11:37,560 Yes. 310 00:11:37,560 --> 00:11:41,360 You shared some articles about how GPUs are being used 311 00:11:41,360 --> 00:11:42,280 for tasks. 312 00:11:42,280 --> 00:11:42,440 Right. 313 00:11:42,440 --> 00:11:43,720 Far beyond gaming. 314 00:11:43,720 --> 00:11:44,080 Yes. 315 00:11:44,080 --> 00:11:45,240 Absolutely. 316 00:11:45,240 --> 00:11:48,840 One area that caught my eye was scientific simulation. 317 00:11:48,840 --> 00:11:49,320 Oh yeah. 318 00:11:49,320 --> 00:11:52,680 It seems like GPUs are being used to model everything. 319 00:11:52,680 --> 00:11:53,000 Right. 320 00:11:53,000 --> 00:11:56,360 From protein folding to the evolution of galaxies. 321 00:11:56,360 --> 00:11:57,280 Amazing, isn't it? 322 00:11:57,280 --> 00:11:59,200 Who thought your gaming rig. 323 00:11:59,200 --> 00:11:59,680 Right. 324 00:11:59,680 --> 00:12:01,920 Could unlock the secrets of the universe. 325 00:12:01,920 --> 00:12:03,120 It's truly remarkable. 326 00:12:03,120 --> 00:12:03,520 Yeah. 327 00:12:03,520 --> 00:12:07,320 Many scientific problems involve crunching massive data 328 00:12:07,320 --> 00:12:07,840 sets. 329 00:12:07,840 --> 00:12:08,280 Right. 330 00:12:08,280 --> 00:12:10,120 And running complex simulations. 331 00:12:10,120 --> 00:12:10,680 Yeah. 332 00:12:10,680 --> 00:12:13,960 GPUs with their parallel processing power. 333 00:12:13,960 --> 00:12:15,000 Yeah. 334 00:12:15,000 --> 00:12:16,480 Excel at these tasks. 335 00:12:16,480 --> 00:12:16,600 Yeah. 336 00:12:16,600 --> 00:12:20,200 They're like supercharged calculators for scientists. 337 00:12:20,200 --> 00:12:22,400 And then there's the world of artificial intelligence. 338 00:12:22,400 --> 00:12:25,680 Which seems almost inseparable from GPUs these days. 339 00:12:25,680 --> 00:12:26,400 Absolutely. 340 00:12:26,400 --> 00:12:28,160 We already touched on tensor cores. 341 00:12:28,160 --> 00:12:28,480 Yeah. 342 00:12:28,480 --> 00:12:30,200 Which are specifically designed. 343 00:12:30,200 --> 00:12:30,480 Right. 344 00:12:30,480 --> 00:12:34,400 For the matrix math that underlies AI algorithms. 345 00:12:34,400 --> 00:12:34,800 Right. 346 00:12:34,800 --> 00:12:36,760 Everything from image recognition. 347 00:12:36,760 --> 00:12:37,120 Right. 348 00:12:37,120 --> 00:12:41,440 To natural language processing relies on these powerful cores. 349 00:12:41,440 --> 00:12:41,800 Yeah. 350 00:12:41,800 --> 00:12:44,920 I know we can't do a full deep dive into AI here. 351 00:12:44,920 --> 00:12:45,240 Right. 352 00:12:45,240 --> 00:12:46,600 That would be a whole other series. 353 00:12:46,600 --> 00:12:48,760 But it's mind blowing to think. 354 00:12:48,760 --> 00:12:49,640 It really is. 355 00:12:49,640 --> 00:12:54,200 That the same technology powering immersive game worlds. 356 00:12:54,200 --> 00:12:54,560 Yeah. 357 00:12:54,560 --> 00:12:59,360 Is also driving advancements in robotics, self driving cars. 358 00:12:59,360 --> 00:12:59,840 Absolutely. 359 00:12:59,840 --> 00:13:01,640 And talentless other fields. 360 00:13:01,640 --> 00:13:05,600 It really speaks to the versatility of the GPU architecture. 361 00:13:05,600 --> 00:13:08,040 It's no longer just about pretty pixels. 362 00:13:08,040 --> 00:13:08,520 Yeah. 363 00:13:08,520 --> 00:13:11,160 It's about solving some of the most complex and challenging 364 00:13:11,160 --> 00:13:13,000 problems humanity faces. 365 00:13:13,000 --> 00:13:13,400 OK. 366 00:13:13,400 --> 00:13:14,880 Before we get to philosophical. 367 00:13:14,880 --> 00:13:15,360 OK. 368 00:13:15,360 --> 00:13:18,440 There's one more crucial component we need to discuss. 369 00:13:18,440 --> 00:13:18,560 Great. 370 00:13:18,560 --> 00:13:19,560 Graphics memory. 371 00:13:19,560 --> 00:13:19,760 Yes. 372 00:13:19,760 --> 00:13:20,320 Graphics memory. 373 00:13:20,320 --> 00:13:21,800 We mentioned it briefly earlier. 374 00:13:21,800 --> 00:13:22,080 Right. 375 00:13:22,080 --> 00:13:26,280 But let's dive deeper into what makes GDDR memory tick. 376 00:13:26,280 --> 00:13:29,280 It's the unsung hero of the GPU world. 377 00:13:29,280 --> 00:13:29,760 Yeah. 378 00:13:29,760 --> 00:13:31,440 Without fast and efficient memory. 379 00:13:31,440 --> 00:13:31,720 Right. 380 00:13:31,720 --> 00:13:34,280 All that processing power would be bottlenecked. 381 00:13:34,280 --> 00:13:34,560 Right. 382 00:13:34,560 --> 00:13:35,840 Starved for data. 383 00:13:35,840 --> 00:13:38,360 The articles you shared talk about GDDR6x 384 00:13:38,360 --> 00:13:39,920 being the current state of the art. 385 00:13:39,920 --> 00:13:40,440 That's right. 386 00:13:40,440 --> 00:13:42,080 What makes it so special? 387 00:13:42,080 --> 00:13:43,640 It's all about speed and bandwidth. 388 00:13:43,640 --> 00:13:44,240 OK. 389 00:13:44,240 --> 00:13:47,480 GDDR6x uses a wider memory bus. 390 00:13:47,480 --> 00:13:47,920 Right. 391 00:13:47,920 --> 00:13:50,680 Meaning more data can flow between the memory 392 00:13:50,680 --> 00:13:52,320 and the GPU simultaneously. 393 00:13:52,320 --> 00:13:52,760 Right. 394 00:13:52,760 --> 00:13:53,200 OK. 395 00:13:53,200 --> 00:13:55,880 And the memory chips themselves operate 396 00:13:55,880 --> 00:13:58,560 at incredibly high clock speeds. 397 00:13:58,560 --> 00:14:00,880 Further boosting data transfer rates. 398 00:14:00,880 --> 00:14:05,600 So it's like having a wider highway with faster cars. 399 00:14:05,600 --> 00:14:06,520 That's a great analogy. 400 00:14:06,520 --> 00:14:08,480 Ensuring a steady flow of data. 401 00:14:08,480 --> 00:14:08,800 Yes. 402 00:14:08,800 --> 00:14:11,120 To the GPU's processing units. 403 00:14:11,120 --> 00:14:11,720 Exactly. 404 00:14:11,720 --> 00:14:13,200 And as we mentioned earlier. 405 00:14:13,200 --> 00:14:13,360 Right. 406 00:14:13,360 --> 00:14:15,680 It also uses those advanced encoding schemes. 407 00:14:15,680 --> 00:14:15,880 Yeah. 408 00:14:15,880 --> 00:14:16,600 Like pay-am-three. 409 00:14:16,600 --> 00:14:19,480 Like pay-am-three to squeeze even more data 410 00:14:19,480 --> 00:14:21,080 through the same physical connections. 411 00:14:21,080 --> 00:14:21,800 That's right. 412 00:14:21,800 --> 00:14:24,360 Now even with all this advanced technology. 413 00:14:24,360 --> 00:14:24,800 Mm-hmm. 414 00:14:24,800 --> 00:14:26,720 There are still trade-offs and challenges. 415 00:14:26,720 --> 00:14:27,360 Of course. 416 00:14:27,360 --> 00:14:27,720 Always. 417 00:14:27,720 --> 00:14:28,840 Memory capacity. 418 00:14:28,840 --> 00:14:29,840 Power consumption. 419 00:14:29,840 --> 00:14:30,800 Heat dissipation. 420 00:14:30,800 --> 00:14:31,080 Right. 421 00:14:31,080 --> 00:14:33,200 These must all be carefully considered. 422 00:14:33,200 --> 00:14:34,000 You're absolutely right. 423 00:14:34,000 --> 00:14:34,640 Yeah. 424 00:14:34,640 --> 00:14:36,560 Designing a high performance graphics card 425 00:14:36,560 --> 00:14:38,640 is a delicate balancing act. 426 00:14:38,640 --> 00:14:41,320 Engineers are constantly pushing the boundaries 427 00:14:41,320 --> 00:14:42,320 of what's possible. 428 00:14:42,320 --> 00:14:42,960 Right. 429 00:14:42,960 --> 00:14:45,760 But they have to work within the constraints. 430 00:14:45,760 --> 00:14:46,400 Right. 431 00:14:46,400 --> 00:14:49,040 Of physics and manufacturing processes. 432 00:14:49,040 --> 00:14:52,480 It's like a high stakes engineering puzzle almost. 433 00:14:52,480 --> 00:14:53,040 Yeah. 434 00:14:53,040 --> 00:14:56,400 Trying to optimize for speed, efficiency, and cost. 435 00:14:56,400 --> 00:14:57,440 It really is. 436 00:14:57,440 --> 00:14:59,880 I'm guessing that's why we see such a variety of graphics 437 00:14:59,880 --> 00:15:00,800 cards on the market. 438 00:15:00,800 --> 00:15:01,680 That's exactly right. 439 00:15:01,680 --> 00:15:03,720 Each tailored to different needs and budgets. 440 00:15:03,720 --> 00:15:04,520 Absolutely. 441 00:15:04,520 --> 00:15:08,080 From entry level cards for casual gamers. 442 00:15:08,080 --> 00:15:08,720 Right. 443 00:15:08,720 --> 00:15:11,760 To behemoth GPUs for professional workstations. 444 00:15:11,760 --> 00:15:12,320 Right. 445 00:15:12,320 --> 00:15:14,240 There's a whole spectrum of options out there. 446 00:15:14,240 --> 00:15:14,640 Yeah. 447 00:15:14,640 --> 00:15:17,920 And the technology keeps evolving at a rapid pace. 448 00:15:17,920 --> 00:15:18,600 Yeah. 449 00:15:18,600 --> 00:15:20,200 So what's considered high end today. 450 00:15:20,200 --> 00:15:20,720 Right. 451 00:15:20,720 --> 00:15:22,000 Might be mid-range tomorrow. 452 00:15:22,000 --> 00:15:22,400 Yeah. 453 00:15:22,400 --> 00:15:24,960 It's an exciting time to be following 454 00:15:24,960 --> 00:15:26,680 the world of graphics cards. 455 00:15:26,680 --> 00:15:27,200 It is. 456 00:15:27,200 --> 00:15:27,800 It is. 457 00:15:27,800 --> 00:15:29,840 They're not just about gaming anymore. 458 00:15:29,840 --> 00:15:30,520 Not at all. 459 00:15:30,520 --> 00:15:33,080 They're driving innovation. 460 00:15:33,080 --> 00:15:33,440 Yeah. 461 00:15:33,440 --> 00:15:34,440 In countless fields. 462 00:15:34,440 --> 00:15:34,840 That's right. 463 00:15:34,840 --> 00:15:36,360 And speaking of the future. 464 00:15:36,360 --> 00:15:37,400 Yeah. 465 00:15:37,400 --> 00:15:39,200 We haven't even touched on some of the cutting edge 466 00:15:39,200 --> 00:15:40,120 research happening. 467 00:15:40,120 --> 00:15:40,600 Yeah. 468 00:15:40,600 --> 00:15:42,280 In areas like neuromorphic computing 469 00:15:42,280 --> 00:15:43,440 and quantum computing. 470 00:15:43,440 --> 00:15:44,120 OK. 471 00:15:44,120 --> 00:15:46,760 Both of which could have a profound impact. 472 00:15:46,760 --> 00:15:47,360 Wow. 473 00:15:47,360 --> 00:15:49,280 On the evolution of GPUs. 474 00:15:49,280 --> 00:15:51,360 That sounds like a whole other deep dive. 475 00:15:51,360 --> 00:15:51,920 It does. 476 00:15:51,920 --> 00:15:52,360 Doesn't it. 477 00:15:52,360 --> 00:15:52,960 Yeah. 478 00:15:52,960 --> 00:15:54,520 But before we get ahead of ourselves. 479 00:15:54,520 --> 00:15:55,080 OK. 480 00:15:55,080 --> 00:15:59,560 Let's wrap up this exploration of current generation 481 00:15:59,560 --> 00:16:00,440 GPUs. 482 00:16:00,440 --> 00:16:00,680 Right. 483 00:16:00,680 --> 00:16:04,120 And Ponder what they tell us about the future of computing. 484 00:16:04,120 --> 00:16:05,440 I'm looking forward to that discussion. 485 00:16:05,440 --> 00:16:05,760 Yeah. 486 00:16:05,760 --> 00:16:07,160 The advancements we're seeing today 487 00:16:07,160 --> 00:16:08,440 are just the tip of the iceberg. 488 00:16:08,440 --> 00:16:09,160 Welcome back. 489 00:16:09,160 --> 00:16:13,640 You know for the final part of our graphics card deep dive. 490 00:16:13,640 --> 00:16:16,040 It's been quite a journey. 491 00:16:16,040 --> 00:16:16,840 It really has. 492 00:16:16,840 --> 00:16:20,600 We've gone from dissecting the hardware 493 00:16:20,600 --> 00:16:25,280 to exploring how GPUs are transforming fields. 494 00:16:25,280 --> 00:16:27,760 Like far beyond gaming. 495 00:16:27,760 --> 00:16:28,120 Yeah. 496 00:16:28,120 --> 00:16:30,120 It's incredible to see how far they've come. 497 00:16:30,120 --> 00:16:30,440 Yeah. 498 00:16:30,440 --> 00:16:31,840 And I think what's really struck me 499 00:16:31,840 --> 00:16:34,800 is how much of this technology is still evolving. 500 00:16:34,800 --> 00:16:35,440 Absolutely. 501 00:16:35,440 --> 00:16:39,040 We're not just talking about incremental improvements. 502 00:16:39,040 --> 00:16:39,600 Right. 503 00:16:39,600 --> 00:16:44,440 It feels like we're on the cusp of some truly groundbreaking 504 00:16:44,440 --> 00:16:45,040 advancements. 505 00:16:45,040 --> 00:16:45,960 I totally agree. 506 00:16:45,960 --> 00:16:49,200 The line between a GPU and a general purpose processor 507 00:16:49,200 --> 00:16:50,440 is blurring. 508 00:16:50,440 --> 00:16:51,040 Right. 509 00:16:51,040 --> 00:16:52,840 You know we're seeing GPUs being used 510 00:16:52,840 --> 00:16:56,240 for scientific simulations, machine learning, 511 00:16:56,240 --> 00:16:57,360 even cryptography. 512 00:16:57,360 --> 00:16:57,760 Yeah. 513 00:16:57,760 --> 00:16:59,400 It's becoming like a fundamental tool 514 00:16:59,400 --> 00:17:00,720 across so many disciplines. 515 00:17:00,720 --> 00:17:03,600 One area that really piqued my curiosity 516 00:17:03,600 --> 00:17:04,560 and the research you shared. 517 00:17:04,560 --> 00:17:04,880 Yeah. 518 00:17:04,880 --> 00:17:07,680 Is the growing role of AI. 519 00:17:07,680 --> 00:17:07,920 Right. 520 00:17:07,920 --> 00:17:09,200 In graphics processing. 521 00:17:09,200 --> 00:17:09,640 Yeah. 522 00:17:09,640 --> 00:17:12,080 It's not just about tensor cores anymore. 523 00:17:12,080 --> 00:17:12,520 No. 524 00:17:12,520 --> 00:17:16,480 It's about using AI to fundamentally change 525 00:17:16,480 --> 00:17:18,160 how games look and feel. 526 00:17:18,160 --> 00:17:18,640 Exactly. 527 00:17:18,640 --> 00:17:21,840 Imagine a game where the lighting, the shadows, 528 00:17:21,840 --> 00:17:26,360 and the textures are dynamically adjusted in real time 529 00:17:26,360 --> 00:17:27,480 based on your actions. 530 00:17:27,480 --> 00:17:28,320 That would be incredible. 531 00:17:28,320 --> 00:17:30,760 You know AI powered character animation. 532 00:17:30,760 --> 00:17:31,200 Yeah. 533 00:17:31,200 --> 00:17:33,560 That's indistinguishable from reality. 534 00:17:33,560 --> 00:17:34,080 Right. 535 00:17:34,080 --> 00:17:35,480 These aren't just pipe dreams anymore. 536 00:17:35,480 --> 00:17:35,880 Yeah. 537 00:17:35,880 --> 00:17:38,240 These are active areas of research and development. 538 00:17:38,240 --> 00:17:41,720 So instead of relying on pre-baked animations 539 00:17:41,720 --> 00:17:44,480 or limited physics simulations, games 540 00:17:44,480 --> 00:17:47,640 could become truly dynamic and responsive. 541 00:17:47,640 --> 00:17:48,120 Exactly. 542 00:17:48,120 --> 00:17:49,440 Almost like interactive movies. 543 00:17:49,440 --> 00:17:53,200 And on the development side, AI could assist artists 544 00:17:53,200 --> 00:17:53,840 and designers. 545 00:17:53,840 --> 00:17:54,600 Oh, absolutely. 546 00:17:54,600 --> 00:17:56,720 Automating tedious tasks and allowing 547 00:17:56,720 --> 00:17:58,880 for more creative freedom. 548 00:17:58,880 --> 00:18:01,000 I think that's one of the most exciting aspects of it. 549 00:18:01,000 --> 00:18:03,800 It really is an exciting time to be working in this space. 550 00:18:03,800 --> 00:18:04,160 It is. 551 00:18:04,160 --> 00:18:04,600 It is. 552 00:18:04,600 --> 00:18:07,200 But wouldn't all this AI processing 553 00:18:07,200 --> 00:18:11,040 require even more powerful GPUs? 554 00:18:11,040 --> 00:18:12,880 Well, it's a classic feedback loop. 555 00:18:12,880 --> 00:18:13,400 Yeah. 556 00:18:13,400 --> 00:18:16,800 As software developers create more demanding applications. 557 00:18:16,800 --> 00:18:17,160 Right. 558 00:18:17,160 --> 00:18:19,600 Hardware engineers are pushed to create more powerful 559 00:18:19,600 --> 00:18:20,760 and efficient GPUs. 560 00:18:20,760 --> 00:18:21,480 Yeah. 561 00:18:21,480 --> 00:18:23,920 And those advancements, in turn, 562 00:18:23,920 --> 00:18:26,560 unlock new possibilities for software. 563 00:18:26,560 --> 00:18:29,200 So it's like a constant arms race. 564 00:18:29,200 --> 00:18:29,920 In a way, yes. 565 00:18:29,920 --> 00:18:32,840 Between software demands and hardware capabilities. 566 00:18:32,840 --> 00:18:33,600 You could say that. 567 00:18:33,600 --> 00:18:35,480 So what are some of the key areas 568 00:18:35,480 --> 00:18:38,280 where we're likely to see breakthroughs? 569 00:18:38,280 --> 00:18:38,800 OK. 570 00:18:38,800 --> 00:18:41,840 One major area of focus is memory technology. 571 00:18:41,840 --> 00:18:42,320 OK. 572 00:18:42,320 --> 00:18:45,720 As we discussed, GDDR6X is the current standard. 573 00:18:45,720 --> 00:18:46,960 Right. 574 00:18:46,960 --> 00:18:49,120 But there are already newer technologies, 575 00:18:49,120 --> 00:18:50,840 like high bandwidth memory or HPM. 576 00:18:50,840 --> 00:18:51,320 Right. 577 00:18:51,320 --> 00:18:51,840 Thanks, Ria. 578 00:18:51,840 --> 00:18:55,800 Nimbroging HPM offers significantly higher bandwidth. 579 00:18:55,800 --> 00:18:57,040 Wow. 580 00:18:57,040 --> 00:19:00,720 But it's also more complex and expensive to manufacture. 581 00:19:00,720 --> 00:19:03,200 So it's a trade off between performance and cost. 582 00:19:03,200 --> 00:19:05,720 Always, as with most things in engineering. 583 00:19:05,720 --> 00:19:07,040 Yeah, but I'm curious. 584 00:19:07,040 --> 00:19:07,400 Yeah. 585 00:19:07,400 --> 00:19:09,560 Are there any alternative approaches 586 00:19:09,560 --> 00:19:11,760 to increasing performance? 587 00:19:11,760 --> 00:19:12,800 That's a great question. 588 00:19:12,800 --> 00:19:16,320 That go beyond just cramming more transistors onto a chip. 589 00:19:16,320 --> 00:19:16,640 Right. 590 00:19:16,640 --> 00:19:18,520 Or pushing memory speeds higher. 591 00:19:18,520 --> 00:19:19,480 There certainly are. 592 00:19:19,480 --> 00:19:19,880 OK. 593 00:19:19,880 --> 00:19:22,800 One area that's garnering a lot of attention 594 00:19:22,800 --> 00:19:24,640 is neuromorphic computing. 595 00:19:24,640 --> 00:19:26,240 Neuromorphic computing, OK. 596 00:19:26,240 --> 00:19:30,240 Which aims to mimic the structure and function 597 00:19:30,240 --> 00:19:31,200 of the human brain. 598 00:19:31,200 --> 00:19:31,880 OK. 599 00:19:31,880 --> 00:19:34,320 Instead of relying on traditional transistors 600 00:19:34,320 --> 00:19:37,480 and logic gates, neuromorphic chips 601 00:19:37,480 --> 00:19:41,200 use artificial neurons and synapses 602 00:19:41,200 --> 00:19:46,160 to process information in a more parallel and energy 603 00:19:46,160 --> 00:19:47,000 efficient way. 604 00:19:47,000 --> 00:19:49,600 So instead of brute force calculations, 605 00:19:49,600 --> 00:19:52,840 it's more about mimicking the way our brains process 606 00:19:52,840 --> 00:19:53,800 information. 607 00:19:53,800 --> 00:19:54,280 Right. 608 00:19:54,280 --> 00:19:57,280 Like pattern recognition, learning, and adaptation. 609 00:19:57,280 --> 00:19:57,840 Wow. 610 00:19:57,840 --> 00:19:59,760 That sounds almost like science fiction. 611 00:19:59,760 --> 00:20:02,800 It is a radical departure from traditional computing 612 00:20:02,800 --> 00:20:03,640 architectures. 613 00:20:03,640 --> 00:20:03,960 Yeah. 614 00:20:03,960 --> 00:20:05,240 And it's still early days. 615 00:20:05,240 --> 00:20:05,720 Right. 616 00:20:05,720 --> 00:20:07,680 But the potential is immense. 617 00:20:07,680 --> 00:20:08,160 Yeah. 618 00:20:08,160 --> 00:20:10,760 Especially for tasks like artificial intelligence 619 00:20:10,760 --> 00:20:12,000 and machine learning. 620 00:20:12,000 --> 00:20:12,320 Right. 621 00:20:12,320 --> 00:20:15,280 Where current GPUs, while powerful, 622 00:20:15,280 --> 00:20:17,720 can still be quite energy hungry. 623 00:20:17,720 --> 00:20:20,080 And I know we've barely scratched 624 00:20:20,080 --> 00:20:21,760 the surface of quantum computing. 625 00:20:21,760 --> 00:20:22,320 Oh, yeah. 626 00:20:22,320 --> 00:20:25,040 Which seems to defy the rules of classical physics 627 00:20:25,040 --> 00:20:25,800 altogether. 628 00:20:25,800 --> 00:20:26,680 It really does. 629 00:20:26,680 --> 00:20:30,040 Could that have implications for the future of GPUs? 630 00:20:30,040 --> 00:20:33,560 Quantum computing is kind of a wild card. 631 00:20:33,560 --> 00:20:34,040 Yeah. 632 00:20:34,040 --> 00:20:36,400 It operates on principles that are fundamentally different 633 00:20:36,400 --> 00:20:37,760 from classical computing. 634 00:20:37,760 --> 00:20:38,280 All right. 635 00:20:38,280 --> 00:20:41,680 While it's unlikely to replace traditional GPUs anytime soon. 636 00:20:41,680 --> 00:20:42,200 Yeah. 637 00:20:42,200 --> 00:20:45,480 There's potential for hybrid systems 638 00:20:45,480 --> 00:20:47,680 that combine the strengths of both approaches. 639 00:20:47,680 --> 00:20:51,920 It's mind boggling to think about the possibilities 640 00:20:51,920 --> 00:20:54,360 we could be talking about solving problems that 641 00:20:54,360 --> 00:20:57,800 are currently intractable for even the most powerful 642 00:20:57,800 --> 00:20:58,680 supercomputer. 643 00:20:58,680 --> 00:20:59,680 Absolutely. 644 00:20:59,680 --> 00:21:03,600 It makes you wonder what the gaming landscape will even 645 00:21:03,600 --> 00:21:05,840 look like in 10 or 20 years. 646 00:21:05,840 --> 00:21:06,080 Yeah. 647 00:21:06,080 --> 00:21:09,080 Will we even need physical graphics cards in the future? 648 00:21:09,080 --> 00:21:09,720 Right. 649 00:21:09,720 --> 00:21:14,760 Or will processing power become so ubiquitous and accessible 650 00:21:14,760 --> 00:21:16,920 that it's just seamlessly integrated 651 00:21:16,920 --> 00:21:18,120 in our everyday lives? 652 00:21:18,120 --> 00:21:22,480 Maybe we'll be streaming photorealistic game worlds 653 00:21:22,480 --> 00:21:24,120 directly to our neural implants. 654 00:21:24,120 --> 00:21:24,960 Who knows? 655 00:21:24,960 --> 00:21:26,400 It's an exciting time. 656 00:21:26,400 --> 00:21:26,960 It is. 657 00:21:26,960 --> 00:21:28,360 To be thinking about these things. 658 00:21:28,360 --> 00:21:29,680 But I think one thing's for sure, 659 00:21:29,680 --> 00:21:32,760 the journey of the GPU is far from over. 660 00:21:32,760 --> 00:21:34,520 I wholeheartedly agree. 661 00:21:34,520 --> 00:21:35,020 Yeah. 662 00:21:35,020 --> 00:21:39,720 It's a testament to human ingenuity, our constant drive 663 00:21:39,720 --> 00:21:41,600 to push the boundaries of what's possible. 664 00:21:41,600 --> 00:21:44,480 And I, for one, can't wait to see what the future holds. 665 00:21:44,480 --> 00:21:45,560 Me too. 666 00:21:45,560 --> 00:21:48,080 Thanks for joining us on this incredible deep dive 667 00:21:48,080 --> 00:21:49,360 into the world of graphics cards. 668 00:21:49,360 --> 00:21:50,240 It's been a pleasure. 669 00:21:50,240 --> 00:21:53,760 We hope you've enjoyed the ride as much as we have.