Behind the Scenes with AMD's Eric Demers - An Interview with Rage3D

Company: AMD
Author: James Prior
Editor: Charles Oliver
Date: July 11th, 2011

The Future is Fusion

Eric Demers, Corporate Vice President and CTO of the Graphics Division at AMD, graciously gave Rage3D's James Prior a few minutes of his time at the AMD Fusion Developer Summit to discuss Fusion APUs, next generation graphics and the future of gaming graphics. This interview took place a couple of hours after Eric delivered the closing keynote address at AFDS presenting AMD's Graphics Core Next Gen architecture, with long time ATI PR manager and all-around good guy Dave Erskine, also in attendance.

James Prior, Rage3D: I'd like to talk about three things, one is Fusion, one is graphics, and one is software.

Eric Demers: Awesome.

AMD Mainstream APU - Llano

R3D: Let's start off with Fusion. The big thing right now is Sandy Bridge, everybody is talking about Sandy Bridge performance, because of the CPU cores. That's really not that relevant for the average consumer; how are you going to message 'Don't care about CPU benchmarks, care about platform experience?'

Eric: The reality today is that CPU is important, but CPUs from 2004 probably solve most of the problems that the average user does, even the office worker of today - if you use Office 2010 pushing the envelope on CPU is not a requirement. So whether it's running at 1.6GHz or 2.5Ghz or 2.9GHz, generally the user experience is very good. Now, on the media side, everybody is about media. Everybody is capturing on these little cameras, everybody is taking pictures; we're not taking bad pictures, in fact it's hard to find a phone with less than 4m pixels, some are pushing 8mp, and any of these little [video] cameras are pushing much more than that. On video it's millions [of pixels], it's HD video we're capturing; media rich is here. People want to go on the web, see HTML5, they want to have a Web 2.0 experience. All of these experiences are in media and media manipulation. They want to play games, whether it's angry birds or whether it's Crysis 2 or DiRT 3, fundamentally we can do things for all of them to make them better and more power efficient and, you know, it'd be great if you could play these games on a mobile platform for a long time. That's what we [AMD] are going to focus on.

Eric: Media Richness, Entertainment Software, Video Playback; all of these things, and those are not about CPU performance. Now there is a little CPU performance [needed] but we offer more than enough to handle that in our Llano cores, and so I think from a user standpoint we deliver exactly what they want. I think it's going to be very popular. And yes, on a PC, CPU only benchmark, they [Intel] have a great set of cores, no doubt about it, they have great and powerful x86. But, the reality is, for most users, that makes no difference whatsoever but the media richness makes the difference. The ability to play games makes a difference, the ability to manipulate video and do it in a very power efficient way makes a difference. Those are the things we'll keep on pushing, hopefully it'll work. Having said that, I'm sure Intel will sell a bunch of those chips as well [laughing], we're just gonna try to take a big chunk out of that because we think we offer a better platform.

R3D: The Rage3D faithful really don't get engaged until the Radeon 5700/6700 level performance and above graphics.

Eric: Right, what I'd call the Enthusiast level, $149 and above, kinda thing.

R3D: Yeah, now Llano is deliberately mainstream but do you see that is going to be a future in enthusiast level graphics or is it always going to stay mainstream?

Eric: We've come to the conclusion that the enthusiast level and above, for the foreseeable future (and the future obviously changes), is for us going to be answered with discrete GPU. We're going to continue to drive, and continue to have updates and great product targeting; actually, the mainstream level and above, because we also have to have discrete GPUs on Intel as well as on AMD. Obviously on the AMD platform we'll have synergistic abilities like dual graphics which presented the ability to shut off the GPU when using the APU and be good about power management. When we detect we have an APU and a GPU that are both AMD we can actually both run in an enhanced mode and give you higher performance than discrete GPU by itself. Those are all things we're going to promote for the AMD platform, but it'll work fine in an Intel platform just like any other graphics will.

Eric: You're right; enthusiast level and above, especially at $199 kind of street prices and above, for now is going to continue to be a discrete GPU. To really be able to offer an APU in that realm would require a much beefier APU and it would somewhat difficult; you know, some of these cards now are pushing upto 225W and in fact some of the higher end ones go beyond that - I would call that ultra-enthusiast, right? [grin] $600+, 6990 kinda thing, I'm not sure on the pricing right now.

R3D: $739

Dave: Most of them are aboot $700, yeah.

Eric: That's pushing 300, more watts; building an APU that delivers that kind of experience would require an APU that has that same size of power profile and has the memory bandwidth associated with it. It would be a [much] bigger APU than we have right now. Now, it is feasible to construct that? Yeah, for sure, it is. Will it happen one day? You know, we're not going to comment on the future, right now we're going to discrete GPU and the APU. This seems to be the right approach but in the future, who knows? Future might be more than a couple years.

R3D: So the future then could be saying to OEMs, 'Hey, you've been happy with a platform TDP of 100W for our CPU/APU and 150W for an add-in board; why we don't we just say that budget just applies to the platform?'

Eric: Yes, that's correct. And that's exactly what we do on the mobile side, where we look now. Historically it's been: this is a 45W platform, and it's got - I'm gonna make up some of these numbers a little bit - but let's say it's got a 30W CPU budget, and it's got the remaining 15W (and that 15W is actually a pretty typical GPU wattage constraint), and so in that platform you just separate them.

Eric: But the reality is both, even going forward, we can do this power sloshing on the APU and move back and forth. There's no reason you couldn't do some of that between the GPU and a discrete GPU, or a discrete GPU and an APU. All of those things are going to happen in the platform but you're right - the APU just looks at the whole thing for now, and that's what we solve to. That gives a lot more flexibility in terms of what it can do [compared to existing CPU + dGPU or iGPU].