AMD discusses 64-bit and dual core at GDC 2005

Processor manufacturer AMD has had great success in the gaming market this past year thanks to the Athlon 64, but new processor technologies are on the horizon. Intel and AMD will both introduce new dual core processors later this year, and Microsoft is getting ready to release its 64-bit Windows operating system which will finally give developers a reason to support 64-bit applications.

We met with Jonathan Seckler AMD senior product manager for the microprocessor business group at this year's Game Developers Conference to get an update on how AMD is approaching 64-bit desktop computing and the upcoming multicore processor revolution.

GameSpot: What does AMD hope to accomplish here at GDC?

Jonathan Seckler: Our goal at GDC is to reach out to the developers and continue to evangelize 64-bit computing on Windows, Windows XP 64-bit. As you probably heard last week, Microsoft has mentioned that April is their announcement timeframe for XP-64 Edition. We think that this is a great platform for the enthusiast market, especially for a 3D game. The games that we've seen running 64-bit, like Shadow Ops, are pretty compelling in the sense that you've got this wider field of vision, a wider, more-realistic world that they're building in these engines now. And so we think that's all goodness. We're also here to emphasize that the FX platform, the Athlon 64 FX, is still the best processor for 3D and high-end gaming. And we don't see that ending anytime soon.

GS: We've been hearing a lot about dual-core processing as the next great advance in processor technology. Isn't that going to help gaming?

JS: Dual core is very sexy. It's a great new innovation for computing. Unfortunately, the enthusiast community, the guys who play these games, is actually going to be disappointed. The games aren't ready for it yet--the environment's not ready for it. When you slow the clock speed down in order to get two cores on a chip, game performance is going to suffer.

So we look at dual core as something that we definitely want to bring to market in the FX platform, but not right away. We need to work with the gaming developers to get their games developed, and that's another reason we're here--to continue that discussion with a lot of the developers. Before we come out with a gaming FX that's dual core, we want to make sure there are games and software.

GS: Let's say a year and a half from now, will there be more triple-A games with 64-bit support, multithreaded support, or both?

JS: Oh, both. I definitely think both. I think in the next 18 months, you're going to start to see a lot of 64-bit and multithreaded games being brought to market. I think that the standards for these games are being raised. We've got a great presentation in our booth. I don't know if you've had the chance to see it, yesterday or today, but we've had a couple of guys come out and talk about how they're developing 64-bit [games].

Basically, you're setting the bar higher. You know how you have to develop games to a base level so that--so you get the [compatibility] exposure. With 64-bit computing, it's becoming such a pervasive thing. They're able to set the bar so much higher and develop to a higher standard. And I see a world where, 2006, everything we make will be 64-bit.

GS: What exactly does 64-bit bring to gaming? Is it allowing a game to do more at the same speed versus having a game do the exact same thing, but faster?

JS: I think you've hit the nail on the head. What we found with the 64-bit games that are out there now is that after, realistically, after 30 or 40 frames per second, you can't see a difference from a video standpoint. So a lot of these guys design the game to run at their optimum frame speed, frames per second, and what they're doing with 64-bit is now taking advantage of the architecture to do more.

It's running at the same speed, but in the case of [a game] like Shadow Ops, the flames are more real. Or when the physics are more real in the sense that things break, they break realistically. When they blow up, it's not just fire, it's smoke, the palm trees moving, and things like that, that you don't have the resources to do when you're in 32-bit. I think on the server side you'll be able to connect more users into the game. Like at a LAN party, if you were running an Athlon 64 as a server, instead of having five guys on the server, maybe you can have 10 or 20.

GS: From the Intel dual-core information released at the Intel Developers Forum last week, it looks like Intel plans to create its initial dual-core processors by grabbing two perfect, adjacent cores from a wafer and slapping them down onto a chip. Are you guys doing the same thing?

JS: No, we're not doing two independent cores. What they're doing essentially, it's like buying a Hershey bar and just breaking off two of the squares at once. You've got to hope that the two squares are perfect, right? We looked at it from a manufacturing standpoint--you create this grid of chips, and this one might run at so many gigahertz, and the one right next to it might run slower. Or this one might run at the rated speed and at the rated voltage, but this one might need more voltage. It's really hard to say, "OK, I'll take those two and they'll work." So they'll be throwing away a lot of chips.

GS: How will AMD design its dual-core silicon?

JS: Our design here, that's the Athlon 64 today. You've got a CPU, all the registers and units and etc. You've got some arbitration logic, and then you've got an integrated memory controller and HyperTransport technology there. All we're doing is adding a second core to the design. And there's space already there. There's already a second port on the system request interface ready for the CPU. There's already a crossbar. The arbitration logic was already built in, the micro code was already there, and it was really designed in from the start.

This dual core die is no larger really than the dies that we have today, from a chip size standpoint, because this will be a 90-nanometer die. Currently the FX is 130 nanometers, so you shrink it to get to 90 nanometers, then you add the core back, and it probably goes back to the same relative size. For us, from a manufacturer's standpoint, it's a very elegant solution.

GS: Do you imagine this will give you a large cost-reduction advantage over Intel?

JS: It's going to give us a cost-reduction advantage. The other thing, too, is it gives us a huge performance advantage. Just think of it conceptually. They've got two dies, they've got the two Hershey bar squares on there. Then they've got to put something else, some other piece of logic to arbitrate, so that when an application comes in it says, OK, die 0, die 1, die 0, die 1, etc. I haven't seen it, but I can't wait to get my hands on it. They'll have a package, they'll have some little pieces of silicon, and they'll have the two die cut together and stuck on the package, and that's how it'll look, right?

We're not going to have that. The way things happen is that requests come in, and this system request interface will arbitrate, and the caches are going to communicate. The communication for the CPU to check to see if this cache is using the information or not, that's all at the speed of the processor. There's no front-side bus. It's not going across the front-side bus and checking and then coming out and back. It's right there. It's natively there, and it's going to be a huge performance benefit, and that's on top of our natural AMD 64 performance benefit. I think Intel is going to be grossly disappointing. I think for anyone who's buying it for what they think they can get out of it, they're not going to get what they expect. And I think with our case, they're going to get more than they expect.