AMD On Graphics: 360 and Wii GPU Guru Speaks

Both the Xbox 360 and the Wii have graphics processing units designed by AMD. After the recent roundtable discussion on the future of console gaming chips, Gamasutra had a chance to sit down with Bob Feldstein, VP of strategic development at the company, and Jon Carvill, PR for graphics.

The discussion touched on the philosophy behind console GPU development, as well as the present and future of the game industry through the eyes of one of its most important collaborators.

AMD has two clients in this race: Microsoft and Nintendo. Obviously, the differences are pretty big between the Xbox 360 and the Wii. Can you talk about how you perceive those?

Bob Feldstein: So, what -- I have to be careful here, obviously -- so, what I can say here is that it's interesting. I think that the Xbox 360 is really exciting to the real gamers, and I think Nintendo is, too. I think that people are going to have two consoles, now. Real gamers are going to have the Wii and they're going to have the Xbox 360-like thing.

And I think the Wii is extending the market. I really think that now, all of the sudden, we have this new dynamic, where just the ordinary person who is not really adept with a controller, will start to have fun with the games. And, yeah, it is about fun; it's about immersion, losing yourself in the game. So, I think both have a place.

I don't think that either suffers from the other, and I think Nintendo's just going to expand the market. And everybody's going to try to take advantage of that next time. So that, you know, that's kind of it. And, as far as the technology inside of them, they're both interesting. Nintendo's immersive, and it shows us there's lots of sides of being immersive. You, do you have both?

Yeah, I buy everything.

BF: Good. (laughs)

To an extent, the technology drives what the developers are capable of doing with a console, but at the same time I would imagine that you take into account what developers might want to do when you're developing a chip. So, how does that work?

BF: It's really a back-and-forth. Sometimes the developers don't know what -- they know what they want to do, but they don't know really what's available in chip technology. And sometimes they presume "not much"; that they have to take care of everything. So if you really think about these advanced developers, they're used to, in the Intel world, say... multi-threading is a problem really left to them; the hardware doesn't help you much.

Whereas, in GPUs, actually multi-threading has been there forever, since 2002. And the GPUs just make sure that things work. They don't tell you that they have a lot of parallel threads running, they make them up on the fly, and they run. And we want to extend that further, and find the kind of features, the kind of play, the kind of other things besides graphics, that are important to game developers -- I mean, physics, obviously -- and intermix those threads on GPUs, and create the kind of compute engines that are easy for them to use, advantageous.

And this is something, by the way, the console game developers are used to. They're almost their own operating system. You put a game on, it takes care of everything. But, the compute model gets more complicated. I don't think, just like they don't want to do assembly language, they want to do C compiling. I think that they're going to be surprised about the kind of libraries and compilers we supply, to make them make it easier.

But you're going to abstract away the complication of the hardware, and you don't want every developer to have to figure out everything from the very beginning. And that's what they do now -- and they're very smart, and they do do a good job -- but I think that as the computer architecture gets more complicated, more parallel, and does more things, that it's going to want to be abstracted away. They want to be involved in the drama of the game, and the beauty of the game, but not necessarily all the time in the metal of how the processors work in the deep, dark metal.

Right now you'll see a number of companies doing different tools that are licensed for many incredibly different solutions -- but it all comes down to creating a game. What are you guys doing, essentially, in terms of that? You talk about doing some libraries...

BF: I'll say that what I see the future as is a combined CPU/GPU, where the delineations are not there anymore. You combine them into a chip; you have compute elements of both, and jobs get kind of -- threads get kind of put on the resources that are best suited for them. GPU-like resources, vector engines for problems that simulate the real world, graphics, and physics. And for sequential problems, about other parts of this, you know, you've got your typical sequential processor.

So, what I see us doing is math libraries, physics libraries, and collaborating with some of those middleware people we were talking about; to make sure that they have the low-level connections to us, so they don't have to figure things out every time. If we change the architecture, they don't have to redo everything -- we can just change some kind of layer.

So are you talking about right now on the 360? Or are you talking about down the road?

BF: Down the road. Down the road. I think now, the 360 model is that... and I'm talking about the model on the 360, is the developers know it really well. They get more and more familiar with it, and the ins and outs, and they do a better and better job. I think that we're going to have more complicated models in the next generation, and those, I think, might be good to share some of the burden of the really low-level programming, and abstraction of all the stuff that can come out of them.

So AMD already sees the next generation on the horizon.

BF: We do. Not that anybody -- you know, we're not doing it -- but, yeah, I think we're pointing some other, future architectures towards that. Because we do think, unlike the Intel gentleman, I think we see the console as a good place for getting this collaboration, and these applications running really well.

His point is currently being discussed by developers -- is the PC the largest market?

BF: I think it's a very interesting observation. Everything's going to be connected to net. It's going to be your handheld device, your phone, your game console, and your PC. I don't know what the distribution mechanism is going to be. And I don't know what we're going to distribute it on. But your PC is going to be a tough one for the living room -- I don't know if it's going to be your traditional game console, here. It could well be your cell phone.

I perceive it like this -- you alluded to the Wii expanding the market, right? So, we have the 360 gamer, and the Wii gamer... So, there's altogether potential that these things can exist in different, parallel markets.

BF: Exactly. Yeah. I mean, and it depends on where you live, and what you do, and all kinds of things that come into it. And, it's going to take -- what I can tell you is that nothing's going to stay the same in ten years. You know? It's going to take angles that we didn't expect. Now, I am always betting on more portable, rather than fixed. That's going to be...

Jon Carvill: I play the 360 when I'm by myself, but I actually play the Wii with my parents and my kids. If you would've told me that there'd be a console that we'd sit, for an evening, and play with my parents, and my eight-year-old, I would've told you that you were smoking something. I never would've seen that coming.

I recently saw numbers -- they're not necessarily 100% reliable, but the seem quite likely -- that nearly 60% of hardware that was sold in Japan so far this year has been the Nintendo DS.

BF: Yeah. DS has been the most successful thing around. And so, remember that. And every little kid in the airport, or wherever you walk, is playing the DS.

But in Japan the other thing you see is you see middle-aged women playing the DS.

BF: Yes you do.

We're not quite there, here.

BF: No, we're not there, but... a question someone asked me, a few weeks ago, an executive at a company... It could've been an executive for my company, for all I'm concerned about, said, "What's going to bring us 49 year olds to playing games on this console?" And my answer is, "You guys are going to die out."

It's the guys who -- he might be brought to it, but there's not going to be [a significant audience], it's the guys who grew up in 1970 and beyond who are just going to play games. And guys, and girls, and whoever else. And so, for them you might see them accepting a DS, or what have you. I think entertainment is more than television, for people after 1970. And especially, and then after 1980 for, you know, a huge leap.

Yeah. Obviously, this was a technical discussion, but some of the questions that were posed, I thought of them in more artistic terms. There was a brief discussion of the emergence of virtual worlds toward the end of the discussion. People were talking about flat display technologies and heads-up displays. But that made me wonder "What do people want?" Not, "What does technology drive?" I guess the thing that comes closest in my mind that is a consumer product that's upcoming, is Sony's Home. And obviously there's stuff like Habbo Hotel -- I don't know if you're familiar with it on the PC -- which is a pretty simple virtual world, frequented by 13-year-old kids.

BF: Yeah, you're right, and people don't want to be driven by technology, of like, "This lighting over here is a lot, a lot better than the lighting over here, because, you know, some game site says so." They want to have fun playing the game. And that's the drama of the game, as much as the beauty of the art, and all that. It's the whole gestalt, the whole thing that gets you into it. Something has to kinda unlock that door.

And the Wii, sometimes, is that easy joystick unlocks the door to this whole fun thing. Other things can unlock that door, too. Just the, you know, just the drama of the game can make it happen. If you're a good game developer, you may not need to have skinning effects, and all kinds of things; you may just need to have stick figures who just -- and you know, sometimes I watch cartoons that, they're funny, and the art is just lousy, but still you create a whole story around it. And that's what the game is. You create a narrative, you get into that narrative, and it's fun.

So when you guys provide a technical solution for a gaming console, how do you look at it -- in terms of what people are going to be doing with it. Discussing the Cell, IBM's James Kahle was saying that they began the project in 2000 and tried to guess what would happen with games in 2007. Obviously, your window probably wasn't that wide with the 360, but how does that work for you?

BF: Yeah it's almost -- it's related to the question that you asked before, as to how we interact with game developers. How we figured out -- we try to do that kind of marketing thing, where we say, "Hey, where do you want to go? What's interesting to you?" We say, "Where's the technology going?"

My gut feeling is now, since we've moved to shaders, what we try to do is make something that is flexible enough that people can find whatever they want within it. Sometimes it'll go slower than it will if it'd been designed five years hence, or something, but they can find within it -- the programmability now is available.

We make the right paradigms, we make it simpler enough, we provide the libraries to abstract some of that stuff away -- but we make a design that is really accessible to the designer, and is extensible by him, because he makes of it what he wants. And that's a nice thing about where graphics is going; from the 2001, 2002 days, and before, 'til now, 2003 and afterwards, it's become more and more. I think you've got to have a lot of flexibility. Because I can't predict the future, and I'm not smart enough. Maybe there are people who are, but...

So, AMD didn't do the main processor in the 360.

BF: No, it wasn't. It was an IBM PowerPC processor -- or, three IBM processors... and so we did the GPU.

Right. I know you guys did that. But you weren't involved with the main processor.

BF: No. I think that's the thing that you'll see changing next time. Because, like I said, the balance is going to change. You're going to actually mix what a processor is, and what a GPU is, you'll have all the elements in one place, and that's what's going to happen.

Now, this is sort of a leading question, but is that what precipitated the ATI-AMD merger, do you think?

BF: Absolutely. In a simpler case... they talk about "integrated graphics". Are you familiar with that?

You mean in terms of a PC?

BF: In terms of the PC and integrated graphics.

The crappy ones on the motherboard?

BF: The crappy ones on the motherboard. And so now you kind of merge the graphics and the CPU. Why have this gap between the two? Because now they're separate; merge them, make them better, and then, over time, make them the same thing. Instead of having a certain number of ALUs to do your math in the CPU, and a certain number of ALUs to do it in the GPU, hey put 'em together, and make the right balance. Because they're different kinds of things.

And, as the fellow from NVIDIA said -- Josh, there -- they're both capable of solving problems beyond graphics, and both capable of solving graphics. Put 'em together, get the right system balance, and all the sudden have... So this is what precipitated the AMD...

JC: And we're targeted at multiple segments. Consoles, PCs, desktops and handheld devices, digital televisions, and other integrated technologies.

BF: Yes. You'll see it across the whole, yeah. It's fun.

One thing that you said during the roundtable was that the GPU and the technology for the 360 was the "go-forward point". There was a discussion about what's incremental, and what's a new start marker. How does that work for you?

BF: So what happened is, so -- I hope I'm answering your question right -- we came up with an architecture for the GPU for the 360. It was based around this unified shader that itself was a balanced system, that balance -- and I don't want to get into the deep, dark details -- vertex shaders and pixel shaders, and it kind of balanced those into one thing that was identical.

And we decided, hey, we'll put the effort into this for the 360, because we're going to base our future architectures off this. And that's exactly what we've done. So we've announced last year a project, and we're announcing another product soon. I don't know where we are on that announcement...

JC: 360, obviously, we first started talking about the processor at E3, three years ago. And then the first discrete graphics processors for PCs were released into the market that used that unified shader technology were this past May, the HD 2000 series. So it was an interesting thing for us, because really, I don't think people fully understood the processing power that was put into that GPU.

BF: So that's going to be a board for a few years.

And that's obviously the same case on the PS3 side, with both NVIDIA's graphics and with the Cell.

BF: Certainly. Cell is, you know, it's Cell.