Skip to main content

Carmack, Sweeney and Andersson Unplugged

The big three rendering architects effectively did an AMA at the G-Sync launch. Here's the transcript.

The rise of mobile graphics

Nvidia is readying a revolution in mobile graphics. Project Logan - Tegra 5, in effect - marries up quad-core ARM Cortex A15s with its desktop Kepler architecture. In the desktop space, Nvidia combines SMX units (each clustering 192 CUDA cores) together in parallel to scale up performance. Mobile Kepler only features one SMX, but based on desktop performance, this should still easily outclass the graphics chips in the current-gen consoles - even Wii U. So the question is, what will happen when that level of performance enters the market. Will we finally see smartphones and tablets host titles that match the quality of the best in current-gen console gaming?

Press QuestionNext year we have mobile Kepler that Nvidia claims is more powerful than the graphics in PlayStation 3. When do you think it will be viable that we will have games like Gears of War, Battlefield, Rage on a smartphone?
John Carmack

So I'm actually spending most of my time right now working on an Android platform, I mean there's some interesting quirks... I'm able to do some rendering directly to a front buffer on a certain Android platform in a way that I can't on any of the PC stuff. I'm twisting arms to try get some of this addressed right now so I have a lot of real current experience. And, there's still a huge gap where the best mobile GPU now is still less than a tenth of what a pretty decent PC is.

You know, for the last couple years they've all been talking about console level performance and they're still not really there. Even the latest and greatest is not as powerful as the current Xbox 360 but it's a tremendous slope that they're on. There are a lot of other challenges in the mobile space even aside from the raw horsepower like I've been commenting how, if you look at an Nvidia Shield the GPU and CPU are pretty potent but if you play the games on there it still drops some frames and has some hitches and things - and that winds up being due largely to power management on these platforms where you've got this very tight battle between performance and power and it comes off as hitching frame-rates.

G-Sync would be a wonderful thing to have on all those Android platforms. The real question is going to be, obviously, you play different games on mobile than you play when you're sitting down at your desktop or your living room but the prospect of like interplay and docking and there are things about, will we eventually wind up with your only computation device being some mobile thing and with some cloud assets... and there's some cogent arguments, arguments can be made that that might be an optimal sweet spot but as far as seeing the latest Battlefield or whatever on mobile it's still a long ways off.

I mean, it's great that you can do a UE4 demo on mobile [but] it's not like you just compile for mobile on that - but it is extremely potent. I think that, obviously more games are being played on mobile than any other platform, though they're not the 'long term devote your life to them', but more people are tapping on the mobile screens than are tapping on joypads or mouse buttons... It's been very interesting to see the games that have evolved there. Probably we're going to see a slight slowing in the rate of increase in GPU power because they've been trading up die size for a while and we're kind of approaching some limits and we're approaching a lot of power limits and thermal limits based on that.

The current torrent pace of growth will probably slow somewhat but it's still going to have a ton of additional progress and I'm going to be really, really happy if Nvidia can capture some significant market share with essentially Kepler on that because, I know I'm working with some of the mobile vendors right now on some of the GPU stuff, but Nvidia just has the development relations and tool and driver team that's better than anybody else. It's probably better than everybody else in the industry put together and it would be nice to get backend support on some of the mobile GPUs.

"For the last couple years they've all been talking about console level performance and they're still not really there. Even the latest and greatest is not as powerful as the current Xbox 360 but it's a tremendous slope that they're on." - John Carmack, Oculus VR.

Nvidia's Faceworks Ira demo running on the Project Logan prototype, miniaturised down from the original GTX Titan version, which ran on hardware that features 14 SMX units compared to the one used to render this.Watch on YouTube
Johan Andersson

Yeah so, we do a lot of experiments and one of the issues that we're facing a little bit is not necessarily from the technical point of view, we're getting there, and we're practically working on all mobile platforms to, well, to prepare Frostbite 4 to experiment to see what we can do with current architectures and future architectures and do various experiments and porting and adaptation. There are quite a few questions to it, even if you could just play your traditional game that you can play on a console well from a technology point of view, the gaming devices are so massively different, the business model is so different, expectations from the user is so different also so, in our view, we'd like to come up with the different type of game concepts ideas and work on the types of game ideas that we have and make that in the mobile spaces.

Our problem was that we have so many game teams that are trained to be working on the console and the PC mindset for such a long time and it's difficult to break out and do these really small type of games that still has interesting value. There are lots and lots of stuff that can be done and I think we'll probably start seeing a little bit of a hybrid type of game, some games that are, perhaps, little bit more of the traditional console experience on tablet, for example, perhaps not cellphone as they're too small and some of those will be really awkward and some may fit really well.

There will probably be this type of Halo moment when you found that, OK, I can actually play a first person shooter with a gamepad but we need that for tablets and there will be probably a couple of them and not specifically target the first person shooters, although that is interesting as well, but likely be for different type of genres. We have that, 'ah-ha, this works' moment. OK, you don't just need to do birds with your single finger, that's pretty low bandwidth input device but you can do more, but I don't know what all those Halo moments will be. But that's real exciting, actually, and having the technology powerful enough both on the GPU and the overall CPU and hopefully also memory systems that's usually a major major bottleneck on mobile and even down to platforms.

You simply have the convenience but you don't have the bandwidth but it feels like we're getting into the trajectory there where it should not be the hardware's fault, it should be our fault if we can't figure out cool ideas and game concepts well enough. So hold us accountable.

Tim Sweeney

Yeah, I think Nvidia coming in as a player on mobile devices is going to become a really important thing for the whole industry. Of course it's a great hardware architecture, it brings DirectX 11 features there and that's all much needed but Nvidia's unsung heroes are the quality of its drivers and this is something that is really lacking throughout the rest of the industry. That quality is there but the performance overhead is really severe. You know, a lot of platforms we're losing a factor of three or four in performance that we could get just in the driver overhead. So we've been taking some of the next-generation game prototyping work that we've been doing and the only mobile platform we're able to run it on is Nvidia's new hardware that they're working on. Mobile targeted hardware that has a driver that is super efficient that can actually run, you know, a game developed for PC and console platforms and adds a driver that's fast enough to do that.

The hardware in a lot of the other devices is fully capable of that level of GPU performance but they're losing a lot in software and Nvidia has played a great role in pushing everybody in the right direction there and we're very happy to see it work and really looking forward to seeing increased competition on that front. So important would be not to just look at the GPU performance, but you have to look at how that's balanced, all of the other components in the system and that's where the real major opportunities are. It's improving the whole system level performance by optimising all of the parameters and making sure they're all tuned against each other.

"We've been taking some of the next-generation game prototyping work that we've been doing and the only mobile platform we're able to run it on is Nvidia's new hardware that they're working on." - Tim Sweeney, Epic Games.

Another old Nvidia demo, first seen running on GTX 480, comes to the Project Logan platform. Tessellation is a high-end feature not usually found on mobile - Kepler's support for the full-blooded Open GL graphics API makes this possible.Watch on YouTube
Press QuestionAre CPU power and memory bandwidth expanding quickly to accommodate a GPU architecture like Kepler?
John Carmack

I used to say like, in the old days, it was always exciting talking to Intel coming down, talking about new stuff as we were getting important things like FPUs and 32-bit address space and things like that but it's been years since I've actually really cared because they're doing a good job, they're chugging along giving us better processing at lower power and it seems like it's one of those things that I just don't have to pay attention to anymore because it seems pretty adequate. I don't think there's a lot left on the table that somebody's just not exploiting on the CPU side. I mean, yes, I'm dearly looking forward to the day when we get some, you know, quantum nano technology whatever that gets us 50GHz CPUs but I'm not holding my breath for it because I think it's, you know, proceeding at pace right now.

Tim Sweeney

Yeah, I think memory bandwidth seems to be in a good balance as long as you're using your mobile device as it's built but once you start beaming your mobile scene to a large display device like a television then you're going to want a whole lot more. Because a lot of the techniques we're using, you know, for deep framebuffers, for high dynamic range, and really advanced compositing techniques just aren't translatable quite yet to a mobile platform. They really work technically, it's just a bandwidth limitation. It's going to be really interesting. Bandwidth costs translates to power right now so these new technologies like DICE, that game, as well as, you know, new memory architectures are going to be really interesting.