Processor Test 1: AI
Designed to focus on a game’s artificial intelligence (AI) algorithms (essentially 3D path finding), this test involves competition among numerous airplanes, each of which slaloms through gates, avoiding collisions. Technically, this test is, of course, entirely parallelized (one thread per core) and should run perfectly on systems with multi-core CPUs. The processors won’t provide a more fluid frame rate, but will calculate much better trajectories for the planes, with the score expressed in the number of operations per second, corresponding to the number of paths calculated for the planes during the test.
Though this test also uses the same 3D engine as the preceding ones, the impact of the GPU is reduced as much as possible by the absence of post-processing (except the tone mapping indispensable for the HDR-only engine), complex shaders and shadows. It also uses a simplified geometry. So, during our tests with the different graphics cards, we noted only a maximum performance difference of 2.6% on the overall CPU score, based on the two CPU tests.
We now see what the performance was like based on the frequency and number of cores.
Unsurprisingly, unlike currently available games (we’re speaking here in terms of overall performance), this test is more sensitive to the change from four to two cores than to a 33% reduction in clock frequency. The scores here are almost exactly proportional to the number of cores and the frequency - the QX6850 scored 100% higher than the E6850, and a processor clocked at 3 GHz was 43% better than the same processor at 2 GHz.