WEBVTT - The Nvidia Story So Far 0:00:04.120 --> 0:00:07.160 Get in touch with technology with tech Stuff from how 0:00:07.200 --> 0:00:13.640 stuff works dot com. Hey there, and welcome to tech Stuff. 0:00:13.680 --> 0:00:16.160 I'm your host job in Strickland. I'm an executive producer 0:00:16.200 --> 0:00:18.560 with how Stuff Works, and I love all things tech. 0:00:18.960 --> 0:00:21.439 And we left off in our last episode at a 0:00:21.520 --> 0:00:25.200 point when in Video, a brand new company at that point, 0:00:25.320 --> 0:00:28.320 was in bad shape at a shaky start. So if 0:00:28.360 --> 0:00:32.159 you have not listened to the last episode where I 0:00:32.200 --> 0:00:36.120 introduced the concept of graphics chips and cards and in video, 0:00:36.520 --> 0:00:40.080 go listen to that one first. Anyway, In Video's first product, 0:00:40.120 --> 0:00:44.760 the n V one, had not really sold super well 0:00:44.800 --> 0:00:48.519 in the market. It hadn't really captured a whole lot 0:00:48.520 --> 0:00:51.680 of market share, and the company had to cancel the 0:00:51.720 --> 0:00:55.760 in V two chip after it's big partners, Sega, announced 0:00:55.760 --> 0:00:58.280 it was going with a different chip designer for the 0:00:58.320 --> 0:01:03.400 graphics for the upcoming dream Cast console. But the engineers 0:01:03.400 --> 0:01:06.320 at Nvidia rolled with the punches and they went back 0:01:06.319 --> 0:01:09.400 to the drawing board and designed a new graphics chip 0:01:09.480 --> 0:01:13.800 that would rely upon the humble triangular polygon rather than 0:01:13.840 --> 0:01:18.440 the quadrilateralals that the Envy one employed for texture mapping. 0:01:19.040 --> 0:01:22.400 This would make their new card compatible with Windows based 0:01:22.560 --> 0:01:26.480 machines that were running the direct three D a p I, 0:01:27.000 --> 0:01:31.559 and the new card would include polygone texture mapping, which 0:01:31.600 --> 0:01:34.440 reduced the amount of time it took for the graphics 0:01:34.440 --> 0:01:37.360 card to render frames, but it had a negative impact 0:01:37.440 --> 0:01:40.720 on image quality. So what the heck is texture mapping? 0:01:41.080 --> 0:01:45.000 Does that even mean? It might sound like gobbledygook, Well 0:01:45.000 --> 0:01:47.800 it pretty much is what it sounds like. It involves 0:01:47.800 --> 0:01:51.960 taking a two dimensional surface called a texture map. Think 0:01:51.960 --> 0:01:55.000 of it like wallpaper or wrapping paper. Think of a 0:01:55.040 --> 0:01:59.720 pattern on that, and then virtually wrapping a three dimensional 0:02:00.000 --> 0:02:04.280 object with that texture. So it's it's a virtual three 0:02:04.280 --> 0:02:08.040 dimensional object. So it's kind of like taking wrapping paper 0:02:08.360 --> 0:02:11.120 and wrapping it around a box. Let's say you've got 0:02:11.200 --> 0:02:13.360 a box, You've got a gift you're gonna give somebody, 0:02:13.680 --> 0:02:16.160 You have some wrapping paper. You have to take this 0:02:16.240 --> 0:02:19.120 wrapping paper that's in a two dimensional format, that being 0:02:19.160 --> 0:02:21.480 a flat sheet of paper, and wrap it around a 0:02:21.480 --> 0:02:27.480 three dimensional object, that being a rectangular cubicle type box. 0:02:28.280 --> 0:02:30.880 And you have to fold the pattern around the contours 0:02:30.880 --> 0:02:32.400 of the box in order to wrap it, which is 0:02:32.760 --> 0:02:36.120 easier to do with simple geometric shapes, but as you 0:02:36.160 --> 0:02:39.800 introduce more complicated shapes, it gets harder to do very well. 0:02:40.200 --> 0:02:42.760 And it doesn't take very much to do to make 0:02:42.760 --> 0:02:45.000 this really hard. I mean you can if you wanted 0:02:45.040 --> 0:02:47.680 to map a simple two D shape like a pattern 0:02:47.680 --> 0:02:51.120 of squares on a three D object like a sphere, 0:02:51.639 --> 0:02:54.720 you're gonna run into problems because if you took an 0:02:54.720 --> 0:02:58.160 actual sphere, like a physical sphere, and you had a 0:02:58.160 --> 0:03:01.600 wrapping paper that had boxes on it, like let's say 0:03:01.639 --> 0:03:05.000 it's just black and white boxes on it, a checked 0:03:05.000 --> 0:03:09.360 pattern almost, and you wrap that sphere with that wrapping paper, 0:03:09.400 --> 0:03:11.320 you you're gonna find that it's going to require you 0:03:11.360 --> 0:03:13.520 to cut the paper in such a way that some 0:03:13.600 --> 0:03:17.280 point on the surface of the sphere the pattern would 0:03:17.400 --> 0:03:20.400 no longer be continuous. There'd be a break in the pattern. 0:03:20.800 --> 0:03:23.560 The same thing can happen with texture mapping, and it 0:03:23.600 --> 0:03:26.680 means that your graphic quality can suffer as a result. 0:03:26.760 --> 0:03:31.280 It can become noticeable. The political texture mapping strategy did 0:03:31.320 --> 0:03:35.400 require less processing power to cover three D objects with textures, however, 0:03:35.840 --> 0:03:39.280 so you have a trade off. The quality of the 0:03:39.320 --> 0:03:43.000 graphics might not be quite as high, but you can 0:03:43.080 --> 0:03:46.080 render them much more quickly, which could improve things like 0:03:46.200 --> 0:03:49.560 the refresh rate. While in Video was working on this, 0:03:49.880 --> 0:03:52.080 another company came out with a product that had a 0:03:52.160 --> 0:03:55.600 big impact on the graphics card industry, and it also 0:03:55.720 --> 0:03:58.640 led to conditions that convinced me to get out of 0:03:58.720 --> 0:04:03.360 PC gaming for quite some time. Not putting blame here, 0:04:03.920 --> 0:04:08.240 just that it created an environment that I got increasingly 0:04:10.080 --> 0:04:14.160 disenchanted with the company, and their product was not a 0:04:14.160 --> 0:04:16.000 bad one, by the way. The company was three d 0:04:16.320 --> 0:04:21.040 f X Interactive. It launched a year after Video had incorporated, 0:04:21.040 --> 0:04:25.760 so he launched in and they introduced a three D 0:04:25.880 --> 0:04:30.279 accelerated graphics card called the Voodoo. And this card was 0:04:30.360 --> 0:04:33.600 different from other graphics cards that had come before it 0:04:34.040 --> 0:04:37.520 because up to that point, graphics cards would typically integrate 0:04:37.760 --> 0:04:41.640 two D and three D graphics processing on a single motherboard. 0:04:42.040 --> 0:04:46.359 But three d FX created a card solely dedicated to 0:04:46.520 --> 0:04:50.400 three D graphics. So you could not just have a 0:04:50.480 --> 0:04:53.039 Voodoo graphics card and a computer and have that be 0:04:53.120 --> 0:04:56.159 your only graphics card and have that work. You actually 0:04:56.200 --> 0:04:59.279 had to make it work with in parallel with a 0:04:59.360 --> 0:05:02.640 two D graphics card or really in series, because what 0:05:02.680 --> 0:05:04.560 you would do is you would take a v g 0:05:04.760 --> 0:05:08.800 A cable from your video controller card, your your basic 0:05:08.920 --> 0:05:12.160 graphics card in your computer. You would feed that cable 0:05:12.600 --> 0:05:15.680 into your Voodoo card, you know, plug it in on 0:05:15.760 --> 0:05:18.560 the back of where the expansion card is, and then 0:05:19.000 --> 0:05:21.880 you'd take another cable and you'd hook the Voodoo card 0:05:22.000 --> 0:05:26.400 to your monitor, so you effectively had two different graphics 0:05:26.400 --> 0:05:29.720 cards doing work inside your computer. The Voodoo handling all 0:05:29.880 --> 0:05:32.960 the three D stuff, and it gave gamers a new 0:05:33.000 --> 0:05:36.360 option when building out computers to play games like Doom 0:05:36.440 --> 0:05:41.200 and Quake, which were progressively putting more emphasis on three 0:05:41.320 --> 0:05:44.479 D models, and they were pushing the boundaries of what 0:05:44.600 --> 0:05:47.680 PCs could do. At that time. Before the Voodoo, the 0:05:47.680 --> 0:05:51.039 most common way to keep up with gaming developments would 0:05:51.040 --> 0:05:53.880 be you'd go and get an upgrade to your CPU, 0:05:54.120 --> 0:05:57.000 which get it really expensive. So I remember when my 0:05:57.080 --> 0:05:59.839 family went from having a two eight six IBM compatible 0:06:00.120 --> 0:06:02.560 to a four a D six, and then later on 0:06:02.600 --> 0:06:08.080 we got a Pentium. But these graphics accelerator cards meant 0:06:08.200 --> 0:06:11.200 you could just buy a graphics accelerated card for for 0:06:11.360 --> 0:06:13.479 much less money than it would cost to upgrade the 0:06:13.480 --> 0:06:16.840 whole CPU, and you could insert it into an expansion 0:06:16.880 --> 0:06:20.640 slot inside your computer, plug up a few cables, and 0:06:20.680 --> 0:06:22.960 provide the processing of you would need to have really 0:06:22.960 --> 0:06:27.040 great graphics. But it also helped usher in an era 0:06:27.160 --> 0:06:31.480 in which new graphics cards would debut at an alarmingly 0:06:31.960 --> 0:06:36.840 frequent rate and make the previous cards obsolete very very 0:06:36.880 --> 0:06:42.000 shortly after they had debuted. Some were two D cards 0:06:42.040 --> 0:06:44.919 that also had three D accelerators on them, some of 0:06:44.960 --> 0:06:48.520 them were dedicated three D accelerators, like the Voodoo, and 0:06:48.600 --> 0:06:50.880 more games were coming out that would push the limits 0:06:50.960 --> 0:06:53.400 on the fastest processors on the market. So it became 0:06:53.440 --> 0:06:57.320 this vicious cycle where you would have brand new cards 0:06:57.360 --> 0:07:00.360 coming out that were much more capable than the ones, 0:07:00.480 --> 0:07:05.520 so everyone wanted one because you thought, who faster means better, 0:07:05.560 --> 0:07:09.600 more features means better. Developers would start developing games that 0:07:09.600 --> 0:07:13.320 would take advantage of those greater capabilities. So while it 0:07:13.320 --> 0:07:16.240 would seem in the very short term that you were 0:07:16.240 --> 0:07:19.080 getting a head start by getting one of these powerful cards, 0:07:19.440 --> 0:07:23.960 pretty soon developers would end up creating applications that would 0:07:24.000 --> 0:07:26.640 push those cards to their limits, and it just became 0:07:26.680 --> 0:07:30.920 this kind of seesaw approach. And that's when I decided 0:07:31.600 --> 0:07:34.280 I don't want to deal with this, and I stopped 0:07:34.320 --> 0:07:36.760 being a PC gamer for a while, and I switched 0:07:36.760 --> 0:07:40.080 to consoles for for quite some time. I did eventually 0:07:40.080 --> 0:07:43.600 go back to PC gamer, but it was in this 0:07:43.680 --> 0:07:47.679 era of numerous graphics cards, and it weren't they weren't 0:07:47.680 --> 0:07:50.720 just graphics cards too, that you also had sound cards 0:07:50.720 --> 0:07:52.800 that were doing a very similar thing at the same time, 0:07:53.480 --> 0:07:55.600 and it was causing all sorts of issues, things like 0:07:55.680 --> 0:08:00.920 questions about compatibility and affordability. I just too frustrating for me. 0:08:01.240 --> 0:08:04.760 But it was a boom time for PC gamers who were, 0:08:05.160 --> 0:08:08.640 you know, eager to pursue that hobby, and they weren't 0:08:08.760 --> 0:08:11.600 weren't put off by the idea that there was suddenly 0:08:11.640 --> 0:08:14.680 a ton of different products out there now. It did 0:08:14.680 --> 0:08:21.120 not help that by were looking at around seventy companies 0:08:21.160 --> 0:08:24.560 that were marketing graphics cards of some sort. Some of 0:08:24.560 --> 0:08:27.480 them were working with technology that was compatible with the 0:08:27.520 --> 0:08:32.040 open g L standard, many more were developing cards that 0:08:32.080 --> 0:08:36.479 were specifically geared to support Microsoft's Direct three D Application 0:08:36.559 --> 0:08:40.000 Programming Interface, and more than a few had their own 0:08:40.040 --> 0:08:44.200 proprietary application programmer interface that that developers could use to 0:08:44.320 --> 0:08:47.760 create software that would look fantastic if it ran on 0:08:47.840 --> 0:08:51.440 those cards. There was a ton of market confusion, not 0:08:51.520 --> 0:08:54.720 just for end consumers, but also for programmers who were 0:08:54.760 --> 0:08:57.840 making software in the first place. Well a Voodoo from 0:08:57.920 --> 0:09:01.160 three D f X. Actually it would officially debut in 0:09:01.240 --> 0:09:04.960 nine and the following year in Video would come out 0:09:05.040 --> 0:09:08.400 with their second commercial product. This was their follow up 0:09:08.480 --> 0:09:10.480 to the n V one. This was the n V 0:09:10.720 --> 0:09:13.679 three because the n V two was canceled. This one 0:09:13.760 --> 0:09:17.959 was better known as the RIVA on. RIVA or r 0:09:18.040 --> 0:09:21.840 i v A stood for real Time Interactive Video and 0:09:21.920 --> 0:09:25.680 Animation Accelerator and this was a hundred twenty eight bit 0:09:25.960 --> 0:09:28.960 three D processor. That's why the one was in the name, 0:09:29.360 --> 0:09:32.520 and that means the processor and busses could operate on 0:09:32.520 --> 0:09:36.559 one eight bit integers or data units that were sixteen 0:09:36.640 --> 0:09:41.840 octets wide. This gave the RIVA a bandwidth of one 0:09:41.880 --> 0:09:45.199 point six gigabytes per second, which is the same thing 0:09:45.200 --> 0:09:50.680 as twelve point eight gigabits per second. Incredible bandwidth. The 0:09:50.800 --> 0:09:54.000 processor was a big hit with O E M or 0:09:54.160 --> 0:09:58.160 original equipment manufacturer customers. So these were companies that either 0:09:58.240 --> 0:10:01.840 made computers or they may graphics cards and they would 0:10:01.840 --> 0:10:06.640 incorporate the REVA in them. It had four megabytes of 0:10:06.720 --> 0:10:10.960 RAM and a two hundred six mega Hurts processor. Um. 0:10:11.000 --> 0:10:13.720 There was a faster model that came out shortly thereafter, 0:10:13.760 --> 0:10:17.679 called the Revo z X. That one had an additional 0:10:17.840 --> 0:10:20.280 four megabytes of RAM, bringing up to eight and had 0:10:20.320 --> 0:10:23.320 a clock speed of two hundred fifty mega hurts. This 0:10:23.400 --> 0:10:28.319 chip was sold to those original equipment manufacturers, and one 0:10:28.360 --> 0:10:32.800 of them was Diamond, which had also been one of 0:10:32.840 --> 0:10:34.880 the O E M s for the original n V 0:10:35.040 --> 0:10:38.760 one chip. They introduced the Diamond of Viper three thirty 0:10:38.840 --> 0:10:41.080 which had a Reva one chip as its three D 0:10:41.160 --> 0:10:45.360 graphics accelerator, and I read some reviews of the revoight 0:10:45.520 --> 0:10:48.079 from that era. I actually found you can find reviews 0:10:48.120 --> 0:10:50.880 online the date back to when this chip came out, 0:10:51.440 --> 0:10:55.760 because the web was a thing in and the reviews 0:10:55.760 --> 0:10:59.079 were pretty positive. Critics were impressed with the three D performance. 0:10:59.520 --> 0:11:03.960 The review is also said that, you know, obviously the 0:11:04.040 --> 0:11:07.680 quality of the performance didn't come just from in video, 0:11:07.760 --> 0:11:12.480 but also from the manufacturers that produced the other equipment, 0:11:12.520 --> 0:11:16.280 the the graphics cards themselves, like the motherboard for the 0:11:16.280 --> 0:11:18.640 the that the chip was sitting on, so it was 0:11:18.679 --> 0:11:22.520 more than justin video, but in video overall came out 0:11:22.559 --> 0:11:26.520 looking pretty good. From the Reva however, it did create 0:11:26.559 --> 0:11:28.720 a brand new problem for in Video but this one 0:11:28.840 --> 0:11:32.320 was a welcome problem, not a bad one. It turned 0:11:32.320 --> 0:11:33.920 out the company was not able to keep up with 0:11:34.000 --> 0:11:38.400 demand because the Reva was a huge economic and critical 0:11:38.440 --> 0:11:41.480 success for the company, and so in Video would sign 0:11:41.520 --> 0:11:46.920 a manufacturing deal with T s MC to supplement manufacturing 0:11:46.920 --> 0:11:50.840 that was already going on with their established partner. They 0:11:50.920 --> 0:11:55.319 had been manufacturing chips through a company called st Micro 0:11:56.160 --> 0:12:01.040 or Microsystems, and they decided to supplement that and use 0:12:01.080 --> 0:12:05.120 a second second partner to provide even more of these. 0:12:05.559 --> 0:12:09.439 By the end, in Video had gone from a struggling 0:12:09.440 --> 0:12:13.920 company on the virginal collapse to owning twenty fo of 0:12:13.960 --> 0:12:17.480 the market share in three D graphics processors. Not a 0:12:17.520 --> 0:12:21.880 bad turnaround, and as it turns out, they had higher 0:12:21.880 --> 0:12:24.800 heights to climb. I'll say more in just a moment, 0:12:24.840 --> 0:12:27.280 but first let's take a quick break to thank our sponsor. 0:12:35.679 --> 0:12:38.720 At this stage, the market was starting to clear out 0:12:39.200 --> 0:12:42.080 a little bit. Some of the companies that had entered 0:12:42.120 --> 0:12:44.480 the graphics cards space we're starting to go out of business. 0:12:45.040 --> 0:12:48.920 But there were three companies that were essentially dominating the 0:12:49.080 --> 0:12:53.480 high performance graphics cards space. Those three companies were a 0:12:53.600 --> 0:12:58.160 T I, three D f X and Nvidia. Some companies, 0:12:58.200 --> 0:13:01.000 in an effort to survive in this increase singly narrow market, 0:13:01.120 --> 0:13:04.680 began to focus on creating budget graphics cards that had 0:13:05.040 --> 0:13:09.120 fewer features, They had a slightly lower performance level, and 0:13:09.200 --> 0:13:12.160 they were generally of a lower cost, so that would 0:13:12.679 --> 0:13:15.520 put them in the range of a different market for gamers. 0:13:16.080 --> 0:13:19.640 But in video was on pretty solid ground now. They 0:13:19.679 --> 0:13:21.880 wanted to follow up on the success of the Reva 0:13:23.120 --> 0:13:26.720 with their next card, which was the Riva TNT also 0:13:26.760 --> 0:13:30.000 known as the n V four T n T stood 0:13:30.000 --> 0:13:34.119 for twin Texel because this card actually added a parallel 0:13:34.160 --> 0:13:38.800 pixel pipeline. It's a lot of ps there to get 0:13:38.840 --> 0:13:42.199 the pop filter going, but this would double the rendering 0:13:42.280 --> 0:13:45.160 speed for the card, and it was a two D 0:13:45.520 --> 0:13:48.040 and three D graphics card. It had support for thirty 0:13:48.040 --> 0:13:51.600 two bit true color graphics. They had expanded the RAM 0:13:51.679 --> 0:13:54.360 up to sixteen megabytes, and it was available both in 0:13:54.440 --> 0:13:58.000 PC I and A g P formats. Those would be 0:13:58.040 --> 0:14:02.880 the buses that you could use with personal computers, and 0:14:02.920 --> 0:14:05.920 there was two different formats for buses. The A g 0:14:06.080 --> 0:14:08.880 P was the newer one of the two and the 0:14:08.960 --> 0:14:11.680 superior one. As far as performance goes of the two, 0:14:12.160 --> 0:14:15.440 but they in video. This time we're mostly focused on 0:14:15.480 --> 0:14:20.080 PC I simply because a g P had not rolled 0:14:20.120 --> 0:14:22.840 out to a very wide user base yet it was 0:14:22.880 --> 0:14:28.240 still new enough where it was not widely adopted as 0:14:28.280 --> 0:14:32.280 the news standard yet. The one downside to in Video's 0:14:32.320 --> 0:14:35.560 design was that the complexity of the chip they had 0:14:35.600 --> 0:14:38.360 made with the tn T meant they weren't able to 0:14:38.440 --> 0:14:41.920 run it at what they had originally planned with their 0:14:41.960 --> 0:14:45.360 memory core speed. They originally wanted to have a memory 0:14:45.360 --> 0:14:48.000 core speed of a hundred twenty five Mega hurts, but 0:14:48.280 --> 0:14:51.720 they found out that because they had made the chips 0:14:51.720 --> 0:14:54.040 soaked or the card so complex, they had packed it 0:14:54.080 --> 0:14:56.840 with so much stuff that if they tried to run 0:14:56.840 --> 0:15:00.280 it at a Mega hurts, the card would over heat, 0:15:01.480 --> 0:15:04.040 so they were forced to throttle the speed down to 0:15:04.240 --> 0:15:07.840 nine d Mega hurts to avoid those overheating issues. The 0:15:07.840 --> 0:15:10.400 big competitor to the t n T was three D 0:15:10.640 --> 0:15:14.400 f x IS Voodoo Too card, which in benchmark tests 0:15:14.960 --> 0:15:18.160 pretty much dominated the three D acceleration world. But the 0:15:18.240 --> 0:15:21.960 Voodoo Too still required a second two D graphics card 0:15:22.000 --> 0:15:24.840 to work. It was not a standalone all in one card, 0:15:25.400 --> 0:15:28.000 so if you were to add up costs. If you 0:15:28.000 --> 0:15:31.520 were building a brand new PC from the ground up, 0:15:31.920 --> 0:15:34.600 you're buying all the different components. If you went with 0:15:34.640 --> 0:15:37.440 a Voodoo too, the costs actually would go up a 0:15:37.480 --> 0:15:40.840 bit because you had to have a secondary graphics card 0:15:40.880 --> 0:15:45.160 in order to run the Voodoo two, whereas in videos 0:15:45.200 --> 0:15:47.520 T and T wasn't all in one you could just 0:15:47.840 --> 0:15:50.920 buy one card and get all of that, So it 0:15:50.960 --> 0:15:53.400 all depended on what your needs were and what you 0:15:53.480 --> 0:15:57.720 already had at your disposal. Also, behind the scenes, three 0:15:57.800 --> 0:16:00.760 d FX was trying to head off competitors like in 0:16:00.920 --> 0:16:03.880 Vidio and A T I by marketing and selling their 0:16:03.880 --> 0:16:08.120 graphics card boards themselves directly. So three d f X, 0:16:08.160 --> 0:16:10.760 in order to do this, purchased a company called STB 0:16:11.040 --> 0:16:13.600 Systems for a hundred forty one million dollars in the 0:16:13.720 --> 0:16:17.720 stock deal. But it turned out the cost of manufacturing 0:16:17.880 --> 0:16:21.880 ended up being higher with STB Systems foundry than the 0:16:21.920 --> 0:16:24.480 foundries that A T and nine and Video were using, 0:16:24.560 --> 0:16:31.000 So the the actual expenses of producing the boards was 0:16:31.120 --> 0:16:35.840 higher than competitors costs, and so that would eat into 0:16:35.880 --> 0:16:38.840 your profit margin. If you're still trying to price your 0:16:38.960 --> 0:16:44.200 your cards competitively against A T I and Nvidia. So 0:16:44.600 --> 0:16:47.520 three d f x is misstep became in Video's opportunity 0:16:47.600 --> 0:16:50.560 because some of three d f x's former partners began 0:16:50.600 --> 0:16:54.160 to drift over to work within Video instead. But while 0:16:54.160 --> 0:16:57.720 in Video was doing well against his competitors, specifically through 0:16:57.760 --> 0:17:00.600 the t n T card, it was dealing with other 0:17:00.680 --> 0:17:04.200 issues that were not so great. Silicon Graphics Incorporated now 0:17:04.440 --> 0:17:08.800 s g I filed a lawsuit against Nvidia. The company 0:17:08.840 --> 0:17:12.080 alleged that in Vidia had infringed on a patent related 0:17:12.119 --> 0:17:16.000 to texture mapping. That dispute would stretch from the spring 0:17:16.119 --> 0:17:20.439 of n all the way into nine but eventually the 0:17:20.480 --> 0:17:23.080 two companies would reach a settlement, and as part of 0:17:23.119 --> 0:17:25.280 that settlement, in Video would actually get s g i 0:17:25.359 --> 0:17:29.439 s professional graphics portfolio, and they hired on several of 0:17:29.600 --> 0:17:32.800 s g i s low level graphics team members, and 0:17:33.000 --> 0:17:36.520 SGI sort of transitioned out of the graphics business at 0:17:36.560 --> 0:17:40.280 that point and started to focus on other areas of business. 0:17:40.320 --> 0:17:43.480 It never really worked out for that company. The company 0:17:43.600 --> 0:17:46.159 s g I was already reeling from other problems, and 0:17:46.160 --> 0:17:49.240 it would ultimately go into bankruptcy in two thousand nine. 0:17:49.800 --> 0:17:52.720 And Video released the t n T two and the 0:17:52.760 --> 0:17:55.679 t n T to Ultra the following year, which effectively 0:17:56.200 --> 0:17:59.680 rested the title of most powerful three D accelerator away 0:17:59.760 --> 0:18:02.040 from three D f X and the Voodoo three card. 0:18:02.400 --> 0:18:05.800 At least in most bitchmark tests. Graphics cards are weird 0:18:05.880 --> 0:18:10.359 by the way you typically test these by running various 0:18:10.880 --> 0:18:15.240 tests that that check the different settings and different features 0:18:15.320 --> 0:18:17.920 of these graphics cards, and it's not always a one 0:18:17.920 --> 0:18:20.959 to one match. You don't always have features that measure 0:18:21.080 --> 0:18:25.639 up where there's an equivalent on a competing card. So 0:18:25.960 --> 0:18:28.720 depending on the test, you might have one card come 0:18:28.720 --> 0:18:31.159 out on top, and then you change tests and you 0:18:31.280 --> 0:18:33.280 run those same two cards through it, and the other 0:18:33.320 --> 0:18:36.600 card comes out on top. So it's a little complicated 0:18:36.600 --> 0:18:38.920 to talk about, but this was a point where in 0:18:39.040 --> 0:18:43.600 Videous performance was look looked at as best in class, 0:18:43.640 --> 0:18:46.600 at least at that moment. Uh. And also keep in mind, 0:18:46.640 --> 0:18:48.800 we're still talking about a world that had different A 0:18:48.880 --> 0:18:51.000 p I s out there that developers were using to 0:18:51.000 --> 0:18:54.639 build software that had three D graphics in it, and 0:18:54.720 --> 0:18:59.080 so sometimes you would find a program that would run 0:18:59.240 --> 0:19:02.639 really well on a computer that had a three D 0:19:02.760 --> 0:19:05.679 f X card, but not as well on a computer 0:19:05.760 --> 0:19:08.320 that had in video card, and then meanwhile you might 0:19:08.320 --> 0:19:13.400 get a different program made by a different developer that 0:19:13.520 --> 0:19:18.160 the opposite is true. So it's very complicated. On October one, 0:19:18.480 --> 0:19:22.080 and Video would define a new era because it announced 0:19:22.080 --> 0:19:26.399 a new graphics card called the g Force two. It 0:19:26.520 --> 0:19:29.840 previously had the code name in V ten. This was 0:19:29.880 --> 0:19:33.480 the first chip to get the designation of GPU, or 0:19:33.560 --> 0:19:37.880 graphics processing unit. The very first GPU was an Nvidia 0:19:38.640 --> 0:19:42.439 product called the g Force two. So what was different 0:19:42.440 --> 0:19:45.520 about this card? What was it that merited the creation 0:19:45.640 --> 0:19:48.959 of a new term. Well, it had a transformation and 0:19:49.080 --> 0:19:53.680 lighting engine which tackled floating point calculations to figure out 0:19:53.720 --> 0:19:57.880 how to render transforming virtual three dimensional objects and display 0:19:57.920 --> 0:20:01.480 them in a two dimensional representation of the image. Wait 0:20:01.480 --> 0:20:04.720 what all right? This is actually not as complicated as 0:20:04.960 --> 0:20:06.679 I just made it sound. In fact, a lot of 0:20:06.720 --> 0:20:08.760 you probably are way ahead of me. But just in case, 0:20:09.080 --> 0:20:11.760 here's what I actually mean by that. Let's imagine for 0:20:11.800 --> 0:20:15.440 a second that you can pop into the virtual world 0:20:15.680 --> 0:20:18.600 of a computer, kind of like Entron. So you are 0:20:18.640 --> 0:20:21.679 in a virtual world, and in this virtual world you 0:20:21.720 --> 0:20:25.240 can see three dimensional objects. Right, there are three dimensional 0:20:25.240 --> 0:20:29.560 objects all around you. They truly have those three dimensions now, 0:20:29.560 --> 0:20:32.560 and in the computer world, that's all represented by math. 0:20:32.760 --> 0:20:36.960 They are just virtual objects. But let's pretend that we 0:20:37.040 --> 0:20:39.800 are in that virtual world. We can see them, and 0:20:39.840 --> 0:20:43.320 we can see how those objects transform as they move 0:20:43.400 --> 0:20:47.160 through a space and environment. Maybe they change shape. Maybe 0:20:47.200 --> 0:20:50.479 it represents a character in a video game, and as 0:20:50.520 --> 0:20:53.679 the character moves, we see how this shape changes. We 0:20:53.720 --> 0:20:57.520 can see how light from fixed sources can play over 0:20:57.560 --> 0:21:01.360 the surfaces of these three dimensional images. We can see 0:21:01.400 --> 0:21:05.679 where shadows form, where highlights are. Now, let's say that 0:21:05.760 --> 0:21:08.359 we we have these three dimensional images that exist in 0:21:08.400 --> 0:21:11.560 this virtual world, and now it's time for us to 0:21:11.680 --> 0:21:16.040 display those images to the real world. And our our 0:21:16.119 --> 0:21:19.600 method of doing that. Our medium is a computer display. 0:21:19.640 --> 0:21:23.439 But computer displays are not three dimensional. They are two dimensional. 0:21:23.720 --> 0:21:26.040 They have width and length, but they don't have depth. 0:21:26.840 --> 0:21:29.080 And so you have to figure out how do I 0:21:29.200 --> 0:21:33.399 portray this virtual three dimensional image that has stuff like 0:21:33.560 --> 0:21:36.520 lighting effects going on with it in a way that 0:21:36.640 --> 0:21:40.600 makes sense within a two dimensional medium, so that you 0:21:40.680 --> 0:21:43.720 get the idea of what that object, how that object 0:21:43.800 --> 0:21:48.200 is shaped. Before the g fource two six, the CPU 0:21:48.400 --> 0:21:51.920 handled that sort of work, this transformation and lighting work. 0:21:52.280 --> 0:21:55.320 But the CPU was also doing all the other work too, 0:21:55.440 --> 0:21:59.120 besides the other graphics processing, and that meant that if 0:21:59.160 --> 0:22:02.159 you added in too much detail in your game, everything 0:22:02.200 --> 0:22:05.320 would bog down and run super slowly or maybe even 0:22:05.400 --> 0:22:09.240 crash because the CPU would be so busy trying to 0:22:09.640 --> 0:22:13.639 calculate all the information required to display that kind of 0:22:13.680 --> 0:22:17.200 stuff that it couldn't do anything else. But by moving 0:22:17.200 --> 0:22:21.520 that requirement over to the graphics card, the GPU could 0:22:21.560 --> 0:22:24.359 free the CPU up to do other calculations, and the 0:22:24.400 --> 0:22:28.080 bandwidth the GPU was super high, high enough to handle 0:22:28.359 --> 0:22:31.399 a massive amount of data, so as long as the 0:22:31.440 --> 0:22:36.159 processor was fast, it could go through this information very quickly. 0:22:36.359 --> 0:22:38.720 In fact, that g Force two six had a throughput 0:22:38.960 --> 0:22:42.359 that was about five times greater than that of a 0:22:42.400 --> 0:22:47.520 five fifty Mega Hurts Pentium three processor, So for certain 0:22:47.560 --> 0:22:51.520 types of applications, a GPU is going to be much 0:22:51.560 --> 0:22:55.440 better than a CPU. Plus, the GPU had on board memory, 0:22:56.040 --> 0:22:59.560 it included a second set of parallel pixel pipelines that 0:22:59.720 --> 0:23:02.480 meant you had four total that was double the TNT 0:23:02.680 --> 0:23:06.199 design and game design developers now could create games with 0:23:06.280 --> 0:23:08.760 much more detail without fear of causing a total slow 0:23:08.840 --> 0:23:12.200 down on a system. It was a literal game changer. 0:23:12.840 --> 0:23:15.040 The clock speed for the g Force two fifty six 0:23:15.080 --> 0:23:18.480 was technically slower than that of the Riva T and 0:23:18.560 --> 0:23:22.040 T two, so they actually took a step down in 0:23:22.080 --> 0:23:24.879 their clock speed. They didn't make it faster than the 0:23:24.920 --> 0:23:29.520 previous graphics card. However, the improvements they made in the 0:23:29.680 --> 0:23:33.160 architecture meant that the g Force two fifty six still 0:23:33.200 --> 0:23:37.600 performed at at faster speed than the t n T two, 0:23:37.840 --> 0:23:41.119 even though the clock speed was slower. Now, up to 0:23:41.160 --> 0:23:43.840 this point, I've gone through the Nvidia cards more or 0:23:43.880 --> 0:23:46.479 less one by one because they were really important for 0:23:46.520 --> 0:23:49.280 the foundation and success of the company and its rise 0:23:49.359 --> 0:23:52.760 from startup to industry giant. But from this point forward, 0:23:52.800 --> 0:23:55.440 I'm going to be jumping around a bit. Otherwise, all 0:23:55.440 --> 0:23:57.560 I'm gonna be doing is giving you a long list 0:23:57.560 --> 0:24:00.280 of graphics cards and their specifications, and that is not 0:24:00.640 --> 0:24:05.200