AMD Explains Why Polaris is Important for VR
Maximum PC speaks to AMD’s Roy Taylor about VR, and the potential of its new architecture
Virtual reality: Those two words are on the lips of every techie and journalist inside and outside the industry. Yet VR is not without its flaws. We spoke to AMD’s Roy Taylor to try to understand what the big companies are doing to counter the current problems, and exactly what the future holds for this innovative and exciting new field of display technology.
Maximum PC: We’re here with AMD’s Roy Taylor, corporate vice president of Alliances, Content and VR. Roy, I’ve got to ask: The vast majority of our audience is certainly concerned about VR’s uptake—after all, even with pre-sales reaching into the tens of thousands, it’s still chump change for any game developer or mainstream publisher—so, what do you reckon it will take to make VR a success? Roy Taylor: Well, that’s a very good point about the install base and the size of the market. John Peddie Research says that if you look at the minimum spec—a Radeon R9 290 or GTX 970 or above—the total install base of those is 7.5 million units. Now, that was information from toward the end of the year, so the number’s gone up since then, but it’s a problem because, assuming that all of those can be matched with headsets—and that’s going to take a little bit of time—that’s still not that big enough of a market for some of the games publishers to come out and invest. When I say invest, I mean invest in something really, really tent pole. So, what we’ve gotta do is one of two things: Help prime the pumps—you know, invest in the game developer community, so they can make the content now, without waiting for the total available market to grow— and also help grow the total available market, by producing new graphics products, at new lower price points, that will run the minimum specs. And we’re looking to do both. MPC: So, is that what your aim is with Polaris?
RT: Yes. MPC: With the specs so tightly locked for both the Oculus and Vive, is there any room for graphics manufacturers, like yourselves, to step in with their own technologies, say FreeSync displays for the HMDs? RT: Well, there could be, but I know that we don’t have any plans to make a headset. And the reason for that is that we’re aware of so many coming already. There’s more than 40 other headsets on their way. And what we think is going to happen is two things. First of all, there’s going to be an arms race toward higher resolution headsets. Now, we already have a 4K headset—it’s absolutely beautiful, it runs at 120Hz in our labs—so we think there’s definitely going to be a rush toward higher resolutions. At the same time, we’re aware of new headsets coming along, and the quality is very good, and the price points are fantastic. So, there’s a Chinese headset manufacturer called Dee Poon, for example, that makes a good headset; we’re also excited about Sulon, who we’re working with; and Star VR, who has
great headsets, too. What’s i important, h however, ihis that f for those headsets that don’t have controllers there is a healthy, vibrant controller market, so we’re very pleased with the work that Sixense is doing. We don’t have any direct interaction with Sixense, because it makes controllers, but we think that its success is important to the overall growth of the VR market. So, we don’t need to make our own headset, we don’t want to compete with the people we’re partnering with, and we don’t think it’s necessary. MPC: Last year saw the announcement of Razer’s OSVR. Can you tell us how AMD is involved in championing this new standard of open-source technology? What does it mean for the VR playing field? RT: Yeah, we think OSVR is important, because it’ll allow a common playground for independent headset manufacturers and independent controller manufacturers to come together, so we applaud them for the work they’ve done. We’re supporting them, we’re part of their association, and I’m very pleasedl d to tell ll you thath theyh are integrating our Liquid VR into OSVR, and we’re giving them all of the tools that they need for them to do that. MPC: What challenges do graphics manufacturers face when it comes to the demanding fps resolutions of headmounted displays? How is AMD overcoming these problems? RT: Well, there are a lot of problems to be solved. You know, if you look right now at where we’re at, the current performance of VR is around 2K per eye, at around 10ms of latency, and around 90fps. That’s roughly equivalent to about 8 TFLOPS of performance, which is roughly comparable to our current highest-end graphics cards, and those of our competitors. What we want to get to is 16K per eye, at around 144fps, and zero latency. That’s roughly equivalent to 743 TFLOPS, or 81 times more performance than we have now. So, we’ve got a lot of work to do to keep building bigger, faster, high-performance GPUs, with world-class drivers, and nextgenerationi software—LiquidVRf Li idVR 2.0, LiquidVR 3.0—to keep making the creator’s job easier. MPC: Is that something that Polaris’s 14nm FinFET process will be able to manage—or at least put us on the right path toward? RT: Yes, Polaris has been designed from the outset with VR in mind. You’ve gotta remember that asynchronous compute, which is on our GPUs, was thought of some years ago. We have a brilliant ASIC design engineer called Mike Mantel, and Mike had the idea that we would, at some point in the future, want a GPU that could do multiple jobs simultaneously, without having a traffic jam, so he came up with the concept called asynchronous compute—that was some time ago. Asynchronous compute is absolutely essential for smooth frame rates in VR in DX12, and DX11. MPC: You’ll have to forgive us—we have to ask. Polaris: How is it looking? Is there anythingan you can give us at all on this one? This year’s going to be a doozie for GPUs—I think everyone is excited—so is there anything you can tell us about it? RT: Unfortunately, we’re in our quiet period at the moment. I wish we could provide more but, unfortunately, there’s only everything we’ve already said, and we’re just really excited about it coming to market. We do think the importance of 14nm FinFET cannot be underestimated, though. Current GPUs from ourselves and our competitors are 28nm—when you shrink, the interconnects are closer together, and because they’re closer together, they run faster. Also, because they’re closer together, they use much less power, and you can also get more per wafer. So that’s not telling you anything that hasn’t already been stated, but that’s not to say you can’t put together just what that means. MPC: Thank you very much for your time, that’s very insightful.
RT: It’s been a pleasure.