Shader Performance
Vertex Shader
Here we'll further examine the vertex shader performance for the two Radeon X800 boards we have.
| D3D RightMark - Diffuse + Specular, 3 lights (FPS) | T&L | VS1.1 | VS2.0 | VS2.0 Static Flow Control |
| X800 XT PE | 51.92 | 46.26 | 46.75 | 46.63 |
| X800 PRO | 46.83 | 42.07 | 42.53 | 42.53 |
| 9800 XT | 27.8 | 24.6 | 25.6 | 25.0 |
| 9800 PRO | 25.6 | 22.6 | 23.0 | 23.0 |
| 9700 PRO | 21.8 | 19.51 | 19.7 | 19.7 |
| X800 XT PE % Difference | T&L | VS1.1 | VS2.0 | VS2.0 Static Flow Control |
| X800 PRO | 10.9% | 10.0% | 9.9% | 9.6% |
| 9800 XT | 87.0% | 88.2% | 82.6% | 86.4% |
| 9800 PRO | 103.1% | 104.8% | 103.0% | 102.5% |
| 9700 PRO | 138.2% | 137.1% | 137.6% | 137.1% |
| X800 PRO % Difference | 640x480 | 800x600 | 1024x768 | 1280x1024 |
| 9800 XT | 68.6% | 71.2% | 66.1% | 70.0% |
| 9800 PRO | 83.2% | 86.2% | 84.7% | 84.7% |
| 9700 PRO | 114.8% | 115.6% | 116.1% | 116.2% |
Looking at the Various RightMark Vertex Shader tests we can see that the X800 XT PE is scoring in the region of about 10% greater than the X800 PRO, which is a little better than its theoretical rates may suggest. In comparison to the 9800 XT the X800 XT PE is always about 80% higher in performance, which is a little below its theoretical performance difference.
For the X800 Pro the test perform in the range of 66% - 72% faster than the 9800 XT, which is again just a little below its theoretical difference. It appears that in both cases here the X800 PRO is just falling a little short of where it should do.
| 3DMark03 Vertex Shader (FPS) | 640x480 | 800x600 | 1024x768 | 1280x1024 | 1600x1200 |
| X800 XT PE | 37.1 | 36.8 | 36.7 | 35.9 | 35.2 |
| X800 PRO | 34.1 | 33.8 | 33.6 | 32.8 | 32.0 |
| 9800 XT | 20.1 | 20.0 | 19.7 | 19.3 | 18.8 |
| 9800 PRO | 18.5 | 18.3 | 18.1 | 17.7 | 17.3 |
| 9700 PRO | 15.9 | 15.7 | 15.6 | 15.3 | 14.9 |
| X800 XT PE % Difference | 640x480 | 800x600 | 1024x768 | 1280x1024 | 1600x1200 |
| X800 PRO | 8.8% | 8.9% | 9.2% | 9.5% | 10.0% |
| 9800 XT | 84.6% | 84.0% | 86.3% | 86.0% | 87.2% |
| 9800 PRO | 100.5% | 101.1% | 102.8% | 102.8% | 103.5% |
| 9700 PRO | 133.3% | 134.4% | 135.3% | 134.6% | 136.2% |
| X800 PRO % Difference | 640x480 | 800x600 | 1024x768 | 1280x1024 | 1600x1200 |
| 9800 XT | 69.7% | 69.0% | 70.6% | 69.9% | 70.2% |
| 9800 PRO | 84.3% | 84.7% | 85.6% | 85.3% | 85.0% |
| 9700 PRO | 114.5% | 115.3% | 115.4% | 114.4% | 114.8% |
| % Difference from low to high res. | X800 XT PE | X800 PRO | 9800 XT | 9800 PRO | 9700 PRO |
| 3DMark03 VS | -5.1% | -6.2% | -6.5% | -6.5% | -6.3% |
Unlike the 3DMark2001SE Vertex Shader test, the 3DMaark03 VS test is showing performances much more reflective of the theoretical rates, suggesting there are fewer outside influences on this test than just the graphics performance.
Here we can see that the performance differences between the lowest resolution and the highest resolution are lower on both the X800 XT boards, indicating that the test is slightly fill-rate limited at high resolution and that, despite the lower bandwidth to pixel ratio of the newer boards, the limitation doesn’t appear to be coming from a bandwidth limitation from retrieving the cached vertices.
