David Watters on AMD FirePro™ Graphics


David Watters who recently joined AMD as Head of Industry Alliances, Professional Graphics talks with Steve Waskul about AMD FirePro ISV solutions.

The interview begins with David explaining the new refocus within AMD on with FirePro™ Graphics as a business line. He explains that there is a lot of investment going on with FirePro™ and that AMD is building a team that can go out and work with the ISVs in the various vertical industries as well as with end users to improve their products and ultimately gain market share.

Steve brings up AMD’s recent launch of their FirePro™ W9100 professional graphics adapter at NAB Show. With 5.24 TFLOPS of peak single-precision floating-point performance and up to 2.62 TFLOPS double-precision floating-point performance, Dave explains that this adapter delivers the world’s fastest graphics performance to date. With six mini DisplayPort outputs and support for DisplayPort 1.2 this card can drive up to six 4K monitors using adapter cables. The FirePro™ W9100 also features OpenCL 2.0, Direct GMA and SDI support and is packed with an industry-leading 16GB of GDDR5 memory. Dave also mentions that some systems can be packed with up to 4 of these adapters yielding over 20 teraflops of GPU compute power and the ability to push up to 200 Million Pixels of resolution across 24 UHD 4K displays.

In addition to the FirePro™ W9100 that launched at NAB Show, AMD subsequently launched the FirePro™ W8100 (4.2 TFLOPS and 8GB of GDDR5 memory) and just before the interview AMD announced the balance of the new family (W7100, W5100, W4100 and W2100). So now AMD has new offerings announced that will extend the GPU family all the way down to the entry level.

As the interview continues Dave and Steve talk about AMD’s work with ISVs and how they are able to leverage the large amounts of memory and memory bandwidth available on the new adapters. The discussion touches ray tracing, parallelism, OpenCL 2.0 and how the number of stream processors (2,816 in the FirePro™ W9100) can impact performance of applications like ray tracing. They also discuss how the amount of memory and its speed are critical to the overall performance you can get out of GPUs for parallelizable tasks.

One of the interesting things about the timing of these new adapters is that it comes just as workstation manufacturers are coming out with new architectures that include new CPUs as well. Having all three areas refreshing at the same time doesn’t happen that often and it will provide users with substantial performance gains. Dave covers this and also discusses AMD’s server graphics adapters that can be virtualized and the capabilities of the FirePro™ R5000 which is AMD’s second generation enterprise PCoIP remote multi-display solution that leverages transmission and compression capabilities from Teradici.

The conversation next moves into virtualization and how businesses are virtualizing workstations and desktops.

One of the “fun” parts of the interview is hearing Dave’s enthusiasm for his work and how he seems truly passionate about the capabilities that are now being delivered to end users. Like other folks we’ve encountered it seems like Dave also enjoys his time working on the PC and has enjoyed sharing the creative capabilities it gives with his daughter who at the age of 11 is now having fun designing houses on a mobile workstation. It’s fantastic to hear stories like this where technology is bringing families together working on projects vs. the typical stories of everyone heads down at dinner texting etc.

As a long-time graphics person and as a long-time SIGGRAPH attendee Dave says he’s blown away by what is possible now on the GPU. He relates a funny story about how he wrote a ray tracing program in college that took 45 seconds or longer to render each individual pixel. Just looking at a HD display this would translate into 259 hours or 10.8 days for a single frame. Comparing this to doing “real-time” ray tracing in 4K and having a final image resolved in a matter of seconds is remarkable according to Dave and we agree.

Next the conversation moves on to OpenCL which Dave believes is a game changer. He explains that lots of ISVs have announced OpenCL implementations and that this will allow programmers to access and harness the power that’s being delivered in the GPU within an industry standard framework that’s portable and can run on different GPUs and even different CPUs.

Finally Steve and Dave talk about AMD’s efforts to help developers and ISVs in implementing GPU computing, porting over their applications to OpenCL and working with the wide range of capabilities available to PC gaming enthusiasts using the various AMD graphics adapters. AMD has alliance and engineering teams that support ISVs and they also have a number of industry partners they can connect ISVs to if they don’t want to do the coding themselves. They also have developed the AMD Mantle Graphics API which is a low-overhead graphics API supported by multiple applications, major game engines and publicly available graphics drivers. Mantle provides developers with tools that allow them to directly talk to the GPU with less translation thereby helping to reduce the workload on the CPU and ultimately allowing developers to maximize the capabilities in both the GPU and CPU which can lead to greater efficiencies in code and richer user experiences. Dave explains that a lot of tier 1 titles like Battlefield 4 have implemented Mantle to gain a jump in performance. This has let the high-end gamers turn on more features at higher resolutions in their games which still allowing folks with lower prices systems to still play the games and have decent features turned on.