Geometry Performance
Here we'll use RightMarkD3D to test the vertex shader performance of the X1800's. First we'll look at the performance differences between the boards under different shader profiles.
| D3D RightMark (FPS) | Fixed Function T&L | VS1.1 | VS2.0 | VS2.0 Static FC | VS2.x Dynamic FC |
| X1800 XT | 82.7 | 73.7 | 79.1 | 58.1 | 50.4 |
| X1800 XL | 71.8 | 65.4 | 70.4 | 47.2 | 41.9 |
| X850 XT PE | 57.5 | 52.7 | 53.1 | 47.2 | |
| X800 XT | 56.1 | 51.1 | 51.5 | 44.6 |
| X1800 XT % Faster than: | Fixed Function T&L | VS1.1 | VS2.0 | VS2.0 Static FC | VS2.x Dynamic FC |
| X1800 XL | 15.2% | 12.6% | 12.3% | 23.2% | 20.3% |
| X850 XT PE | 43.7% | 39.7% | 48.8% | 23.0% | |
| X800 XT | 2.5% | 3.1% | 3.3% | 5.8% |
| X1800 XL % Faster than: | Fixed Function T&L | VS1.1 | VS2.0 | VS2.0 Static FC | VS2.x Dynamic FC |
| X850 XT PE | 24.8% | 24.0% | 32.5% | -0.2% | |
| X800 XT | 27.9% | 27.9% | 36.8% | 5.6% |
In these test the previous generation parts generally tend to stay more consistent across each of the different shader profiles with, for instance, only a 10 FPS variance for the X850 XT PE between the tests; the R520 based board, though, have a larger variance between the tests with some form of branching and those that don't. For the tests without branching we see that the X1800 XL is performing very close to it theoretical difference to the X800 XT in the tests, even exceeding it in the VS2.0 test, however this drops off significantly with the static branching tests and there is only a 6% difference between the two. Note that with the X800/X850 only being Shader Model 2.0 parts they have no need to support the optional Dynamic Branching Vertex Shader capability, but with R520 being Shader Model 3.0 based it does support Dynamic Branching, hence it can also be utilised under the extended Shader Model 2.0 profile.
| D3D RightMark (FPS) | Ambient | Diffuse (1 point light) | Diffuse (2 point lights) | Diffuse (3 point lights) | Diffuse + specular (1 point light) | Diffuse + specular (2 point lights) |
| X1800 XT | 136.2 | 138.2 | 142.7 | 103.2 | 128.3 | 76.5 |
| X1800 XL | 125.0 | 125.6 | 116.5 | 83.6 | 117.0 | 64.1 |
| X850 XT PE | 127.0 | 126.6 | 80.2 | 56.9 | 96.3 | 52.7 |
| X800 XT | 131.9 | 131.4 | 74.5 | 53.2 | 98.6 | 49.1 |
| X1800 XT % Faster than: | Ambient | Diffuse (1 point light) | Diffuse (2 point lights) | Diffuse (3 point lights) | Diffuse + specular (1 point light) | Diffuse + specular (2 point lights) |
| X1800 XL | 8.9% | 10.0% | 22.4% | 23.5% | 9.6% | 19.3% |
| X850 XT PE | 7.2% | 9.2% | 77.9% | 81.4% | 33.3% | 45.1% |
| X800 XT | 3.3% | 5.2% | 91.6% | 94.0% | 30.2% | 55.7% |
| X1800 XL % Faster than: | Ambient | Diffuse (1 point light) | Diffuse (2 point lights) | Diffuse (3 point lights) | Diffuse + specular (1 point light) | Diffuse + specular (2 point lights) |
| X850 XT PE | -1.6% | -0.8% | 45.3% | 46.9% | 21.6% | 21.6% |
| X800 XT | -5.2% | -4.4% | 56.5% | 57.1% | 18.8% | 30.5% |
Using the maximum available Vertex Shader profile on each of the board here yields a few interesting results in that in a few cases the R520 based boards are underperforming relative to the previous generation's shader rates, with the X1800 XL displaying a lower performance than the X800 XT in a couple of cases, but other tests display gains greater than their theoretical differences due to the different ways in which the shaders are being executed.
Note: Although Rightmark3D does feature VS3.0 capabilities the X1800's were unable to run the VS3.0 profile for these particular tests so the VS2.0 Extended profile was used for these boards instead. The X1800's do have the VS3.0 Caps bit enabled and are capable of running other tests under the VS3.0 profile in RightMark. The VS3.0 tests here do not have HLSL profiles, instead using hand coded assembly which obviously weren't written on an X1000 board and we suspect that this is where the issue may lie.
