Graphics and Nvidia

processing units. So before we jump into the history of Nvidia and how it helped define the evolution of computers and how it created the term GPU, how about we explore the concept of graphics cards and GPUs in general. A and just a note for you guys, this gets

super technical, super fast. So I am going to take a very high level on all of this because if you really want to know and have a but understanding of the history of GPUs and graphics cards, you also have to have an understanding of the evolution of how we program graphics, things like application programmer interfaces or a p I s um that kind of stuff. You need to know about that evolution in addition to the evolution

of the technology itself. And it gets super technical. And while it might be interesting to a subset of tech stuff listeners. I feel like for a lot of people, I would lose them very quickly. I have a feeling this These episodes are going to lose some folks anyway, and that's okay. But for those who are interested in in video and graphics cards, stick around. We're gonna learn

some interesting stuff. So to understand in video's role, we actually have to dial way back before the founding of Nvidio, which was We're gonna go all the way back to nine teen fifty one, just briefly, because that's when the U. S. Navy awarded a contract to M I T to design a flight simulator, and M I T created a simulator called Whirlwind. That system had an early three D graphics system, but the architecture of Whirlwind wouldn't become the basis for

the graphics card of the future. However, it's important to note that people were thinking about trying to create graphics that would represent three dimensional objects using computer systems as early as the nineteen fifties. But to look at the kind of technology that eventually evolved into the realm where in video got involved, we have to look really more at the birth of Arcades home video game consoles and the personal computer, so this would be the mid nineteen seventies.

Arcade machines and consoles had specialized chips in them, and we typically would refer to them as things like video address generators or video shifters, and the chips or circuits in some cases would take information from a processor and relay that in a meaningful way to a display in order to create the images you would see on a computer monitor or a television. So remember that a processor like a CPU central processing unit, will take in data

from a source. You get input that goes into the CPU. The CPU will perform various operations on that data according to a program running on the machine, and then produce output that is meaningful based upon the combination of the program and the input. And then that output will go somewhere. It might feed into a new process, it might go to an output device, and this is in good old

machine code. So the graphics chips or circuits their purpose was to receive output data from the CPU that was specially scifically intended to go toward a display and then to take it from there. So graphics card act as sort of a translator. They take the data from the CPU, and they turn it into something that's meaningful for a display to show a bunch of points of light on a screen. Remember, the images we see on screens are made up of pixels, and pixels are points of light.

And the smaller the pixels, the more you can fit on a screen, the greater the resolution, just generally speaking. So computers need a graphics card or some sort of graphics processing unit or interface on the computer's motherboard itself or in an expansion card in order to do this. Now, those early chips in the nineteen seventies had names like the c d P eighteen sixty one from our c A or the T I A one A which was used in the ATAR in Motorola debut. The MC six

eight four or five video address generator. That chip give instructions to a display that would include information such as luminants or how bright the image should be, that pixel should be, color, and position. And so the pixels new quote unquote knew how bright they need to be, what color they needed to be, and the display new which pixels needed to be shown it or be activated at

any given time. More importantly for our story, the MC six eight four or five chip for Motorola served as the foundation for the first video card for a personal computer, which was IBMS Monochrome Display Adapter, which was introduced in Now that could not really generate graphics the same way we would talk about today, but it could display text and characters in monochrome initially. Eventually you also had a color one. Intel built a graphics chip called the eight

two seven twenty which could display up to eight colors. Wow, can also support a resolution of two hundred fifty six by two fifty six pixels. Now, if you swopped over to monochrome, you could actually have a better resolution of five twelve by five twelve pixels. And for several years, video game consoles and arcade machines were miles ahead of

personal computers when it came to graphic quality. The circuits in the gaming machines were dedicated to that task, and early arcade games had systems that were essentially hard coded onto circuits, which meant if you wanted to change a game out in an arcade cabinet. Let's say that you've got an arcade cabinet, and inside that cabinet you have I don't know, let's say it's Galica and you want to swap out Galaga for pac Man, Well, you can't just take a chip out and replace it with another chip.

You would actually have to essentially gut the whole cabinet and replace the INNERDS with the new ones from pac Man. There was no swapping a flash drive or anything like that. The advantage of the dedicated systems was that a could really create pretty good graphics, especially compared to the stuff you would see on computers. You get really good results.

But the downside was it was not a viable approach if you wanted to create a general purpose machine like a personal computer, and so PCs for the most part, lagged behind these specialty machines, at least in the graphics department. In the mid nineteen eighties, a company called A t I began to manufacture graphics cards. They would end up

becoming incredibly important in that realm. Their early products were o e M, or original equipment manufacturer cards, which meant they were producing chips that were meant to be included in finished products that other companies were making. So they were they were making a component that would be included in a larger product that some other manufacturer was producing. Um they were not selling directly to the end customers, So you wouldn't go into a store and buy an

a TI chip back in the mid nineteen eighties. Instead you would go and buy a computer that had an a TI chip in it, or maybe a card that had an a TI chip incorporated in it. That would change in the late nineteen eighties. You would start seeing a t I producing its own cards supporting stuff like e G A and v G A graphics and I'm not going to go into those. It would be ridiculously long winded to go into all those different graphics and they largely are unimportant by the time we get around

to talking about in video anyway. So this, however, was one of the big moves that led to the birth of the graphics card industry as a consumer facing business. By the early nineteen nineties, when Nvidio would debut, more companies started making two D graphics cards for computers and three D accelerators typically paired with those two D graphics cards, and then a t I came out with the Mack sixty four that was the the graphics card that also

included support for a TV tuner and full motion video playback. Acceleration, and things were really starting to pick up in those early nineties. Now let's take a step back and consider why this is a big deal. When you're looking at a three D image on a computer screen, meaning one that uses various techniques to simulate depths, a lot has to happen to produce those results. Because what you're looking at is a two dimensional representation of a three dimensional object.

The graphics processing unit has to construct a wire frame for the shape you're looking at, to build out sort of a an exoskeleton, if you will, for the entire scene,

everything from characters in it to backgrounds, et cetera. If it's doing a full three dimensional h field of view, that frame ultimately consists of a bunch of straight lines typically, but the lines can be really short and at angles, and if you have the lines small enough and at slight enough angles from each other, you can simulate curved shapes. And once the wire frame is there, the graphics card raster rises the image, which means it fills in the

pixels kind of like a coloring book. You're filling in the spaces for that wire frame. Next, it adds elements like color and texture and lighting effects. And that's just for one moment, one instance of that image. So if you're playing a game, typically you're having stuff change on screen pretty regularly and sometimes quickly, so the image needs to be updated frequently, and a fast paced shooter, for example,

will not be a fun experience. If your computer can only display a couple of new images every second, you'd be toast. If it were a competitive game, thirty frames per second is considered slow or at least subpar, But if you can get thirty frames per second regularly, tend things tend to be playable. You really want something that's sixty frames per second at least for a smooth experience. And the more graphically rich the image, the more work

that entails. Right, the more detail you have in your in your individual images, the more work your your graphics processor is going to have to do in order to refresh that sixty times a second. So the graphics processing unit has to be able to relay the appropriate information at a bandwidth large enough and a speed fast enough to produce those results, and it's a big job. Graphics

cards typically have four main components. They have a motherboard, which is a circuit board upon which all the other components are on top of UH. The graphics processor and other connections are all sitting on this circuit board, and the motherboard also has a connector to plug into a computer, and that connector allows the motherboard to accept data and power from the computer and send data back to the computer.

Then you've got the processor. It's the processor's job to interpret information from the computer and send data to the display so that the right pixels show up at the right time at the right part of the screen. And the processor specializes in performing geometric calculations and can include features like anti aliasing that's a smoothing technology to take some of those pointy edges off the graphics, since ultimately all the shapes you're looking at are made up of

straight lines. Or there's also an isotropic filtering, which creates a more crisp image. Then there's another component on these cards that's memory. Graphics cards have memory to store information about each of the pixels and temporarily store completed images. UH. Then there's some sort of connector that allows you to plug the display into the graphics card so that way you can actually see the wonderful graphics this card is creating. If you don't have a connector, then there's the display

has no way of receiving that information, right. So also these days you might end up using a connector on the back to plug one graphics card into a second graphics card and run them in series so that you can boost the performance. That is becoming more and more common where you start seeing big gaming rigs that have

multiple high performing graphics cards inside of them. Now, these cards tend to produce a lot of heat because they're working very, very hard, and they've got a lot of components on them, and so they tend to also have multiple cores. That means the processors can work on parallel problems. Most graphics rendering falls into that category, but as it turns out, so do other computational problems, which is why you will hear about large groups using GPUs to tackle

certain types of challenges like mining bitcoins. I'll talk more about that in our second episode, but that's a good basic grounding of graphics cards. When we come back, I'm going to talk about the founding of Nvidia and the environment it found itself in those early days. In nine three, Jensen Huang, Chris Malachowski, and Curtis priam founded in Vidia.

But where did they come from? Well, Huang, who is president and CEO of Nvidia to this day, was born in Taiwan, and then he lived in Thailand as a young boy until his family sent him and his brother to the United States because Thailand was becoming politically unstable, and they moved in with some family members over in the States. His aunt and uncle enrolled Huang in a boarding school in Kentucky. They thought it was a prep school. It turned out that it was actually a school for

troubled youth. He and his brother were reunited with their family and moved to Oregon. Huang would attend Oregon State University. He studied electrical engineering there and graduated with an undergraduate degree. Then he went on to enroll in Stanford University to

earn his master's degree in electrical engineering. After that, he went to work for Advanced micro Devices or a m D. Interesting because we'll hear more about a m D in the next episode, and he also then worked for l s I Logic before he would leave to co found in Video Chris Malachowski, who today is part of in Vidio's executive staff and serves as a senior technology executive.

He attended the University of Florida and received his bachelor's degree there and then went to Santa Clara University for his masters. He went on to work as a manufacturing design engineer at Hewlett Packard in nineteen eighty and in nineteen seven he started working for Sun Microsystems and became a senior staff engineer, and while he was there, he co invented the g X graphics architecture and led design

work on graphics interfaces. Curtis Priam attended rin Alert Polytechnic Institute and he earned a degree in electrical engineering in nineteen eighty two. He worked on the IBM Professional Graphics Adapter and like Malachowski, Prium would end up working at Sun Microsystems and was also one of the inventors of the g X graphics architecture and now the three founders for in Nvidia. He is the one who is no longer working with the company. He retired from in Vidia

in two thousand three. So back in pri Um and Malachowski are steaming. They're incredibly frustrated because they had been encountering roadblocks at Sun micro Systems. They both felt the company was pushing R and D in the wrong direction for graphics, and Wong, who was working at l S I he wasn't really disenchanted by his work, but he was thinking about what the future of processing was going to be all about. And the three sat down to talk things over at dinner at a Denny's restaurant, which

I think is great. The idea that a hugely successful company was born over a conversation at a Denny's is fantastic to me. I've had a lot of conversations at Denny's. None of them have led to a monumentally successful company, so I'm not doing it right. But anyway, the three of them, I started talking about the future of computing, and they agreed that graphics based accelerated computing was the future.

And gosh, they were incredibly prescient about that. We'll we'll talk more about that, I mean, but I kind of alluded to it earlier, about how graphics processing units are now used in lots of different applications that have nothing to do with making pretty pictures on a display. Uh. They kind of saw that coming down the pipe. Maybe not quite in those terms, but they did think that was the future of computing. However, to pursue this belief.

It would mean forming a new company and and leaving their current jobs, and then designing the technology from the ground up, designing graphics chips from the from the beginning, and so it would mean entering into a crowded marketplace because the early nineties already had a lot of different companies that were creating graphics cards, and there were probably

around two dozen competitors already on the market. So it's becoming a huge headache already because software developers had to figure out which formats they were going to support, which a p I s should they use to develop their software that because the APIs would be compatible with specific

types of graphics cards. Graphics cards would support programs that were developed using certain APIs, and you could get things like drivers to help support uh all other API s. But if a company is making a graphics card with the intent to support a specific API, that kind of software is going to perform better on that hardware air than other software. So this was becoming a confusing marketplace for developers as well as for customers. Let's say that

you're out there on the market. You you want to buy a machine that is really graphically intensive so that you can do stuff like maybe maybe you're doing architecture design or or computer assistant design CAD programming. You want to make sure that the systems you buy are going to be compatible with the software that you depend upon. And that was starting to get kind of a little complicated in the early nineties, and it would only get

worse before it would get better. But this environment did lead to opportunities like the founding of companies like alien Ware. You know, alien Ware being the boutique gaming rig company that that all started because the founder of alien Ware would end up building gaming rigs for friends because he was not confused by all the different types of cards and motherboards and things like that out there, and so he would build gaming rigs for them. That was possible

because of this massive confusion in the marketplace. Now, one year before in Video launched, so back in there was another company called Silicon Graphics Incorporated, later known as s g I. They will release a multi platform application programming interface for two D and three D graphics called open g L one. Oh. Now, it was not initially intended to be an a p I for developers to develop three D games or games that would use three D graphics,

but it was quickly adapted for that purpose. And meanwhile Microsoft was preparing a rival API release called Direct three D. And Microsoft didn't concern itself too much to make sure that open g L would run smoothly on Windows machines, which is the kind of thing that would get Microsoft

in trouble down the line. They would get accused of engaging in anti competitive practices, in other words, making sure that their operating systems would support their own software and maybe not do so much to support other people's software. But it also helps illustrate how confusing this world would have become already before the video had even hit the market. But while in Video was forming, other graphics card companies

were already folding or getting absorbed into larger companies. Uh So, in Vidia is coming into a market where companies with names like Gemini Technology, Headland Technology, Tamarack, Acumos, those were starting to go away. Some companies were still around, but they shut down divisions that had been working in graphics processing.

Motorola did that with their division, Acer did that. So in Video was getting into a very uncertain market and the three co founders didn't have any idea how to create a business plan, and they didn't even have a product yet, but they did hold rounds of venture capital financing. To get the startup capital. They needed to design a graphics card and partner with a manufacturer to make it. Because they did not have the manufacturing facility, they were

not actually gonna build the cards. They were going to design them and send those designs to a manufacturer that would be in charge of actually putting the stuff together. Wong has said in interviews that he tried to make a business plan to present to potential investors. He really tried, but he just wasn't able to do it. And essentially he told investors, I tried to make one, but I ran out of time before I could finish it. And the response was essentially, that's okay. We wouldn't have believed

your business plan anyway. But the three engineers knew their technology backward and forward, and they were able to get enough confidence and investors to secure funding to the tune of around twenty million dollars. And I would like to take this opportunity to say that I also do not have a business plan. Money please, I guess it doesn't work for me. Well, anyway, this whole process took about a year, So from ninety three to ninety four, the

company is just kind of getting started. They're starting to design their work, they're raising finances. They form a strategic partnership with a company called s G S Thompson micro Electrics. This company would actually be the one to manufacture the first graphical user interface accelerator chip that in Vidio designed. They also partnered with another company called Diamond Multimedia Systems to take these chips and install them in multimedia accelerator boards.

So if you think of a graphics card like an actual expansion card that you would put in a computer, Diamond Multimedia Systems built the card and the chip that powers the card was made by Nvidia. So producing graphics cards required a lot more work than just one company, is my my point here. The first product out of Nvidia would come to market in n and it was the in V one nd PC. Consumers would know it as the Diamond Edge three D graphics card, and the chip was chosen by Sega to go into the Sega

Saturn game console. The Diamond Edge three D was a conventional pc I card. Pc I stands for Peripheral Component interconnect They would later end up being replaced by other formats, but this was the standard that we used to plug in expansion cards. It was a type of computer bus. A computer bus is essentially a channel or a pathway

that exists between different components and a computer. So PC I card slots would allow PC owners to swap out expansion cards that would allow you to augment your PCs capabilities. It might be a sound card, or a graphics card or something along those lines. So how did Nvidio's first product do well? According to A A Wong, not so great. He said that when they got their finished card back from the manufacturers, they on that of what they had

designed wasn't working in the actual build out. And the card was supposed to process both two D and three D graphics, plus act as a sound card, plus a game pad port for Sega Saturn controllers, plus a PC joystick port, and more so it was packing a ton of features onto a single card. And it might have been a bit too ambitious. Uh. It nearly meant the invidios story could end before it could really begin. But as we'll find out, and video was able to weather

that storm. We'll come back in just a moment, but let's take a quick break to thank our sponsors. Wong has talked about a demo his company held that would show off this en V one ship, and they showed it off at a trade show called Compute Text. He said, even though many of the features were not working in

this demo, it still made a really good impression. But then in Video came back the following year to Compute Text and they showed off the exact same demo, and then he saw that the reactions were not as positive. And that's where he learned that if you show up to the same conference twice, you know, in two different years, and you just show the same demo, that's a big mistake. It sends a bad message. And he said we almost

went out of business. In a piece published in I E. E E. Spectrum that has the title in Video co founders remember their startup Roller Coaster Ride, Huang said quote, we had this idea that computer graphics was going to be the driving force of technology and fuel would be video games. A year and a half later, we nearly went out of business. The technology was wrong, the market strategy was wrong, architecture was wrong, and the execution wasn't

good end quote. So what do you mean the architecture was wrong. Well, for one thing, the way the n V one would render graphics was through a process called quad texture mapping, But just as the cards were starting to ship out of manufacturing facilities, Microsoft was finalizing Direct X one point Oh, that's what incorporated Direct three D and Direct three D did not use quad texture mapping.

It used triangular polygons to map textures. And Microsoft's dominance in the operating systems space for PCs meant the game developers were more likely to build games tailored to Microsoft's approach because most of their customers were Microsoft customers. They weren't gonna go the quad texture mapping strategy that in Video had used, so in Video tried to compensate for this.

They released a driver that would uh if you installed the driver if you had the Envy one chip in your computer, it would try to wrap triangles as if they were quadratic surfaces to give the Envy one a semi compatible platform with Direct three D. But the NV one failed to attract many developers apart from those early Sega ports. Sega obviously had tapped them for the Saturn,

so ports from Sega Saturn to the PC. If you had a in Video V one, they ran pretty good, really good actually, And I think it's interesting that Wong was very candid about this, about how this decision was not a good one early on in the company's history. But Wong and Malachowski have both credited in Videos success at least in part to an acceptance that sometimes things just don't go the way you planned and you have to pivot and make changes. They say their work was

adaptable and that's what allowed them to survive. They could adapt their thinking and their strategies and their technologies to make the best of a bad situation. They could learn from their mistakes and work to avoid them in the future, well,

you know, making brand new mistakes. Despite the issues with the NV one, and Video was able to land a deal with Sega, and ports of Sega games to the PC depended upon that in V one architecture, and n V one had the honor of becoming the first commercial graphics processor that could do three D rendering, video acceleration and graphics user interface acceleration. However, the sales of the NV one were a disappointment, but they did help keep the company above water, so they began work on its

successor the design of the n V two. But then the company got some bad news. Sega had decided they were not going to go within video or uh the graphics chip for their next video game console, which was the Dreamcast. Instead, they were going to go with power VR for that. So in Video had to figure out

what they were going to do next. And speaking of that, in our next episode, we're going to look at how in Video came from this shaky start and we're able to cement themselves as industry leaders in the graphics card arena, how they were able to invent the term graphics processing unit, what that actually means, what is a graphics processing unit, what made their first graphics processing unit a GPU, And we'll also talk about some of the problems the company

has encountered since then. I hope you enjoy this episode in the next one, and if you guys have any suggestions for future episodes of tech Stuff. Maybe it's a technology, maybe it's a company, Maybe it's a person in tech. Maybe there's someone I should interview. Let me know. Send me an email the addresses tech Stuff at how stuff works dot com, or draw me a line on Facebook or Twitter. The handle at both of those is tech Stuff hs W. You can check out our merchandise store

over at t public dot com slash tech stuff. I'm wearing one of those shirts as I sit here recording right now. I'm wearing my Princely sum T shirt. It's so comfy and I look so snazzy in it. I don't even mind that I spent a princely sum on it. You shouldn't either, t public dot com slash tech stuff, and make sure you follow us over on Instagram and I'll talk to you again about Nvidia really soon for more on this and thousands of other topics, because it how stuff works dot com