An interview with ATI - Dave Baumann - Interview with ATI's Dave Baumann - Page 3

Multi-GPU setups are becoming more accepted and popular these days. AMD has a very reputable monthly driver release to support the latest games. I certainly will not mind testing AMD Radeon HD 4890 in a ATI CrossFireX setup. Now, for ATI CrossFireX game compatibility one needs driver support, a driver based game profile. A lot of our readers jumped ship on the ATI CrossFireX solution based on this fact: if a game is not supported in the driver, ATI CrossFireX will not kick in.

When you purchase a new game you want it to be supported right out of the box. Here at Guru3D we have been evangelizing the need for user creatable game profiles in AMD's ATI Catalyst™ drivers. Last year at a summit we heard this was planned. However, up-to this very moment, this feature was never implemented. We've seen several situations where your end-users just bail on ATI multi-GPU solutions, just because they can not make a game profile, select AFR/SFR/etc and enjoy their game in all its glory. Can you get us any insight or updates on this rather peculiar topic?

We take the feedback on ATI CrossFireX game support very seriously and this is something that we are constantly looking at in terms of providing the end user with great ATI CrossFireX scaling as soon as a title is released. The intent is to constantly improve our performance here.

With regards to end user ATI CrossFireX profiling, our software engineers like the idea and feel it can be done, however right now the plan is to make it part of a much wider scope of changes to Catalyst™ Control Center and the driver that may take considerable time to fully implement.

End user profiles are not necessarily the best answer - they may provide benefits to some titles, but they won't work in all cases. At times, we'll still need a driver update to provide the best support. It may be better to provide, effectively, an ATI CrossFireX patch - a small update to the driver that actually applies the correct profile and parameter tweaks in a much smaller update outside of an entire driver revision. This can be delivered outside of the standard monthly ATI Catalyst release, provided much closer to the time a game is released.

Of course, the best solution would be to have the support in the driver even before the title is launched. This is an area that we are focusing on currently. Our ISV team is working diligently to get builds of upcoming titles in-house prior to the game being released so that we can see the types of rendering profiles the app has and can code the ATI CrossFireX profile accordingly; we made some pretty good strides in this area over the past year and early coverage has increased significantly. Additionally, we are working closely with developers to ensure their titles are ATI CrossFireX "aware" from the outset. If you look at recent Direct X 10.1 titles such as Battleforge or Stormrise, they were coded in a ATI CrossFireX friendly manner in the first place.

Finally, you may remember that the drivers have a "Compatible AFR" mechanism by default which was designed to provide some level of scaling in AFR unfriendly titles even without a specific profile. The Compatible AFR mode was implemented in the DirectX 9 era with good results in many DirectX 9 titles. Due to the changes in the rendering path, this has fallen out of compatibility with more recent DirectX 10 titles. Hopefully, soon we will be updating the compatible AFR profile for the DirectX 10 driver in order to better capture more titles straight away.

AMD is introducing a new concept, AMD Sweet Spot GPU strategy. We learned that the concept is based off a simple thesis good performance and features at an affordable price. Could you elaborate a little on that concept, and feel free to evangelize a little.

Obviously the goal is always to provide a compelling price / performance ratio, however beginning with the ATI Radeon™ HD 4800 series we reoriented our design approach to address the market differently, with a laser focus on performance per dollar.

Previously we introduced a new product generation by launching a very large, powerful chip that addressed the enthusiast market segment. Reconfigured derivatives of this high-end part were then used to address the performance, mainstream and value market segments. The issue with this approach was the lag between the first chip and the various derivatives. In some instances, months passed before a new generation of graphics processors had products available for all market segments. This approach also limited our flexibility and agility with respect to reacting to opportunities at a given price band.

Based on research and experience, we surmised that a more effective strategy would be to focus our initial efforts in the $150 to $300 range, a part of the market that had evolved into an incredibl[y]e important, high volume battleground for discrete graphics. When in the planning stages of RV770 the decision was made to specifically target this range in order to gain a lot of traction in the market with the product and then to also very quickly address the high-end enthusiast market with an X2 board solution. This allowed us to address three market segments (performance, enthusiast and ultra enthusiast) in a very short amount of time by using the same ASIC in different configurations.

A key element here was also to be innovative in the design of the product. From the outset, there was a clear goal to scrub the unified architecture from top to bottom and reconfigure it to deliver as much performance as possible at key price points. In the end, despite only an increase of approximately 40% in silicon size over RV670, and on the same process, much of the engine had over 2.5x the number of functional units, a mind boggling improvement! I must mention that I joined ATI/AMD just as the early development of RV770 started and watched it grow from concept to final product - watching it surpass performance milestone after performance milestone pushed me to jump at the chance of managing the product when the opportunity arose.

Of course, with a big engine increase we needed a commensurate increase in bandwidth, which is why we made the decision to utilize GDDR5.

The net result of the sweet spot strategy and the path of innovation have been to offer highly competitive solutions at compelling price points and even now, nine months after the release of the first products, they are still selling very well.

The next topics I like to cover are the move towards DirectX 11 later this year and architectural changes in general. So DirectX 11 will be launched with the upcoming Windows 7. First off, when can we expect DirectX 11 class GPUs from ATI? And what are your feelings towards DirectX 11 in comparison to DirectX 10.1 (how it ties in) and what in your vision will bring DirectX 11 to games.

Right now we are still selling DirectX 10.1 products so I won't go into detail on the timing of DirectX 11 products. As a follower of the GPU industry I'll leave you and your readers to postulate on the development cycles and the timing relative to major inflection points. We should note, though, that given the recent release of several DirectX 10.1 titles, we feel there is still lots of headroom for DirectX 10 and DirectX 10.1. Additionally we should remember that DirectX 11 is a superset of DirectX 10 and DirectX 10.1, meaning that by including DirectX 10.1 developers are already paving the way to DirectX 11 features and compatibility.

By providing support for DirectX 10.1 we're helping developers be ready for DirectX 11 sooner than if they only limited current development to DirectX 10.

As for DirectX 11, naturally we're happy to see the DirectX API continue to evolve in the manner it has. My feelings are that it offers a sensible evolution of the feature-set capabilities, in line with the directions the IHVs are taking from a hardware perspective and where the ISV's want to go on the software side, whilst also addressing some of the points that were lacking in DirectX 10.

One such element that gets updated in DirectX 11 is that of Display Lists, a new driver model to more effectively multithread graphics workloads over multi-core CPUs, natively within the API. This is something that we know developers have been looking requested. The advantage here is that although this is a DirectX 11 API feature, the functionality will move down to DirectX 10 hardware, so all DirectX 10 hardware users that update to the DirectX 11 runtime will get the benefits of this feature.