Synthetic Benchmarks: Vertex Performance

Well, we can’t actually talk about vertex performance of Graphics Media Accelerator 900 itself: in Intel’s scheme all the geometry calculations are performed by the central processor. Intel relies on the might of its CPUs, and we’ve got a fast and very expensive one in our testbed, with 2MB of cache and 3.4GHz clock rate. In the opposite camp, we have ordinary middle and low-range standalone graphics cards on chips from ATI and NVIDIA.

Take note of a curious fact: the results of the graphics cards decline greatly as we switch from one to eight light sources, while the results of Graphics Media Accelerator 900 only go down by a third.

Both tests – with one and eight light sources – have the same number of polygons, and the small performance hit that GMA 900 suffers on turning on eight lights makes me suspect that GMA 900 spends more time on sorting the polygons, rather than on their transformation and lighting – the scene is polygon-rich, and “tying” a polygon to a tile involves a lot of calculations.

GMA 900 coupled with a Pentium 4 Extreme Edition executes DirectX 8 vertex shaders as fast as the GeForce FX 5200 does, at least in this test.

When it comes to advanced DirectX 9 vertex shaders, the Accelerator is twice slower than the slowest graphics cards from ATI and NVIDIA.

Overall, the geometry performance of Graphics Media Accelerator 900 (teamed with Intel Pentium 4 Extreme Edition CPU) doesn’t impress. We also see that the graphics core has problems with polygon-heavy scenes, and they are due to the time-consuming operations of sorting polygons by tiles, rather than due to a low speed of geometry processing by the CPU.

So, GMA 900 has displayed nothing exceptional compared to cheap DirectX 9-comaptible graphics cards in synthetic tests. Let’s see if this will also be so in real games, in comparison to graphics chipsets of the last generation.