Benchmarks

Theoretical Benchmarks

Here we'll take a look at the PowerColor Evil Xabre 400 using 3DMark2001SE's theoretical throughput tests.


 Pixel Fill (Mpp/s) Texel Fill (Mtp/s) Geometry 1 Light (M tris/s) Geometry 8 Lights (M tris/s) Vertex Shader (FPS)
P4 2.53GHz 451.0 1666.0 25.7 7.8 76.6
P4 1.8GHz 438.8 1610.6 25.1 7.7 47.3
% Diff -3% -3% -2% -1% -38%

The first thing that we can see is that there is a large disparity between the single texture and multi-texture fill-rates. With four pixel pipelines and two texture units per pipe, we would expect to see a single texture fill-rate roughly double what is displayed there. Comparing the single to multi-texture rates, it would suggest a configuration of two pixels pipelines each with for texture units.

Looking at the geometry throughput rates we can see that there is hardly any performance variance between processors with the Geometry tests, but there is with the Vertex Shader test. This is consistent with the chip utilising a hardware T&L engine, but not Vertex Shader, as the SiS documentation stated.

SPECviewperf 7.0

We'll take a look at the performance of the Xabre 400 under the OpenGL workstation benchmark. Under some of the SPECviewperf tests there is large quantities of data being passed from the host CPU to the graphics device, so we'll test the AGP4X and 8X modes of the Xabre 400.





3dsmax-01 drv-08 dx-07 light-05 proe-01 ugs-01
AGP4X 6.77 18.40 29.18 9.06 5.83 2.71
AGP8X 7.09 22.80 29.60 9.15 6.78 2.84
% Diff 5% 24% 1% 1% 16% 5%

On the face of it, it would appear that the Xabre 400 has a reasonable level of performance under SPECviewperf, as we've seen even some higher end boards post lower scores than this.

In some of the tests there does also appear to be an appreciable difference between AGP4X and AGP8X modes, with AGP8X proving to be at least slightly better in all cases, and with a 16% increase in the proe-01 test and a 24% increase in the drv-08 test.