As you’ll know if you read our articles regular like, quad-core processors are not yet very well exploited in current games. Although dual core is more and more widespread, very few games can lay claim to making the most of more.
Over the course of various tests designed to observe this employment of quad-cores we came to the conclusion that there was one parameter that had a great deal more impact that expected: the graphics card, or rather, its driver. In our processor test protocol we use two games, the 1.2 version of Crysis (via the integrated benchmark_cpu2), and World in Conflict, version 126.96.36.199 (via the integrated bench). In both cases, the tests are carried out in 800x600 so as to limit the impact of the graphics card and concentrate on processor limitations, with as much load as possible.
Here are the results we got with an E8400 and a Q9650, both clocked at 3 GHz, coupled either with a Radeon 4870 (Catalyst 8.11 hotfix), or a GeForce GTX 280 (GeForce 180.48), in DX9 and DX10 mode (effects are set high in DX10 which is why we get the low frame rate), in 32-bit Vista:
The results speak for themselves! In DirectX 10, with an NVIDIA card, you get between 21 and 24% performance increases by moving from dual to quad core. With a Radeon 4870 the gain is less significant. However in DirectX 9, the gains are smaller: only between 10 and 11% with NVIDIA, and, only in Crysis this time, a more or less identical gain with Radeon. Multithreading is of course not managed in the same way in Vista for DirectX 9 and DirectX 10.
The results obtained show clearly that the DX10 NVIDIA pilots are better multithreaded than the DX10 AMD pilots, which are in fact not at all or very minimally multithreaded. As a result, an NVIDIA card is better able to get more out of a quad core, which is not to be sniffed at if you are playing a game where CPU limitations are compromising performance.
According to Intel, 25 to 40% of the CPU load in a game is linked to Direct3D and the graphics driver. The threading of this load is not therefore insignificant and it would be more than useful for AMD to look at the question as soon as, to at least get on a comparable level with NVIDIA. We are now planning to update our Core i7 test, which we carried out with a Radeon HD 4870 X2 at first, this time using a GeForce GTX 280.