PCi : Why did you choose to use the same three 3DMark05's scenes (although we appreciate they have been improved), rather than creating new ones for the 3DMark06 version ?

TS : Rapid advanced in raw 3D processing power (SLI and Crossfire systems) enabled us to revisit those scenes and raise those to totally different level. We thought that it will be very good way to show in concrete way what a difference you will get with better hardware.


PCi : The CPU benchmarking scene has been made with PhysX API ? What did you exactly implement from AGIEA's technologies ?

TS : 3DMark06 does use Ageia PhysX API library (PCi's note : It is optimised for multi-threaded programing), it is not hooked to take advantage of hardware acceleration, all is done in CPU. We think though that hardware accelerated physics calculations are both immensely important and interesting, from gaming and from benchmarking point of view, as we move forward.

PCi : 3DMark 2006 is optimized for dual core processors. Briefly, how did you manage to take advantage of the dual-core structure? What weighting does it have in the final score ?

NR : We don't have a different weighting for dual-core CPUs vs. single-core CPUs. All CPUs run the same path in 3DMark06's CPU tests, but since the tests are threaded, and the number of threads scales with the number of processors, any dual-core CPU will have an advantage. We worked very hard to get the best benefit out of the dual-cores by organizing and optimizing the CPU calculations (AI, Physics etc.) for parallel execution in multiple threads. We are confident that future games will show similar advantages for dual-cores as 3DMark06 does if the games' CPU code has been optimized for dual-cores/HT.


PCi : What's your developers' opinion on the obvious complexity of coding multi-core optimized applications, especially 3D games ? Do you think it will be even more difficult with Cell's architecture ?

NR : This is an interesting question really, but very difficult to answer. From our experience with dual-cores/HT processors, it is not that difficult to get significant benefit out of them but of course it requires a lot of research and resources to make sure that everything is working optimally also on single core CPUs. I have no doubt that if a game-engine is designed with multi-core CPUs in mind it isn't that difficult to achieve good benefits from dual-core/HT CPUs. Adding support for dual-cores at a later stage in the development is of course trickier. When we designed 3DMark06's CPU tests, we designed them with dual-cores in mind, which means that we spent a lot of time and resources into getting the most out of such CPU's.

We haven't looked at the CELL CPU, so I am not really able to comment on that.

PCi : Some websites tend to stop using your benchmarks arguing that “real”, in-game testing is more relevant. What's your opinion about that ?

TS : It's a free world :) Honestly, if you are a professional reviewer or educated consumer, I am sure that you will want to have as many data points available as possible. 3DMark brings something that today's games can't bring and that's very aggressive workloads (you buy a graphics card today but you most likely will be playing with it for years to come, so why not check how future proof it is). This coupled with the fact that it is a professionally designed product just for benchmarking purpose guarantees the validity of results. We advocate all to use a combination of professional benchmarks and games for game performance benchmarking.


PCi : Will there be some flying cows' tips and tricks after a upcoming release of a hidden code ?

TS: It is our strict corporate policy to not to comment on any possible “easter eggs” that may have found their way into our products :) :)

Page précédente

Page suivante