Benchmarks - Geometry Performance
Here we'll use RightMark D3D to look at the performance of the various geometry shader profiles.
| D3D RightMark (FPS) | Fixed Function T&L | VS1.1 | VS2.0 | VS2.0 Static FC | VS2.x Dynamic FC | VS3.x Dynamic FC |
| 6200 | 46.1 | 17.0 | 17.3 | 10.4 | 13.2 | 13.2 |
| 6600 GT | 69.3 | 30.2 | 30.5 | 18.2 | 23 | 23.2 |
| 5200 Ultra | 22.03 | 6.5 | 6.7 | 3.4 | 4.0 |
| 6200 % Faster Than: | Fixed Function T&L | VS1.1 | VS2.0 | VS2.0 Static FC | VS2.x Dynamic FC | VS3.x Dynamic FC |
| 6600 GT | -33.4% | -43.8% | -43.1% | -43.0% | -42.7% | -42.8% |
| 5200 Ultra | 109.4% | 162.3% | 160.1% | 202.3% | 233.5% |
In terms of comparative performance we see that the 6200's performance differences to the 6600 GT and 5200 Ultra are within the areas you'd expect given the theoretical performance differences - it does appear that the 5200 Ultra has some extra hardware to assist with fixed function processing as its performance difference is lower here than with the shader tests. However, we do see similar trends with the NV4x chips than we've seen before from NVIDIA with fixed function having much higher performance than straight Vertex Shader programs and the performance suffering a little when any kind of branching is utilised.
Pixel Shader Performance
Again, using RightMark D3D we'll take a look at some of the Pixel Shader performances of the 6200 alongside the other boards:
| Rightmark3D, 1024x768 (FPS) | 6200 | 6600 GT | 5200 Ultra |
| PS1.1 Procedural | 147.5 | 549.1 | 156.1 |
| PS1.4 Procedural Procedural | 114.7 | 373.4 | 88.1 |
| PS2.0 Procedural | 78.8 | 283.24 | 37.9 |
| PS2.0 1 Light (FP) | 69.5 | 241.22 | 16.8 |
| PS2.0 1 Light (PP) | 96.3 | 333.77 | 18.8 |
| PS2.0 3 Lights (FP) | 38.9 | 133.69 | 7.6 |
| PS2.0 3 Lights (PP) | 50.3 | 178.4 | 10.0 |
| PS2.x 3 Lights (FP) | 17.0 | 58.6 | 2.8 |
| PS2.x 3 Lights (PP) | 22.7 | 78.37 | 3.6 |
| 6200 % Faster Than: | 6600 GT | 5200 Ultra |
| PS1.1 Procedural | -73.1% | -5.5% |
| PS1.4 Procedural Procedural | -69.3% | 30.3% |
| PS2.0 Procedural | -72.2% | 107.9% |
| PS2.0 1 Light (FP) | -71.2% | 314.4% |
| PS2.0 1 Light (PP) | -71.2% | 411.7% |
| PS2.0 3 Lights (FP) | -70.9% | 415.0% |
| PS2.0 3 Lights (PP) | -71.8% | 401.2% |
In comparison to the 6600 GT we see the 6200's performance difference under these Pixel Shader tests are very much inline with the theoretical texture-rate differences between the boards, which is no surprise as the fragment pipelines in NV43 contain the texture units and so these metrics are tied together.
In comparison to the 5200 Ultra, though, the 6200's performance differences are somewhat different. With the PS1.1 Procedural test the 5200 Ultra is performing faster than the 6200, probably because this is quite reliant on bandwidth and texture performance, which the 5200 Ultra has a slight advantage with. When the shader scale up through the capabilities range the 6200 has a much greater performance than the 5200 Ultra, up to over 5 times the performance in the longer PS2.0 shaders.
ShaderMark V2.1 has the capability for rendering its shader utilising PS3.0 functionality, and here we'll take a look at the performance differences between the rendering on 6200 in PS2.0 and PS3.0. (Note - only those with performance differences are listed)
| ShaderMark V2.1 | PS2.0 | PS3.0 | % Diff |
| shader 3 Per Pixel Directional Light Shader (Phong) | 134 | 135 | 0.7% |
| shader 6 Per Pixel Anisotropic Lighting | 129 | 130 | 0.8% |
| shader 13 Per Pixel Veined Marble Shader | 73 | 77 | 5.5% |
| shader 15 Per Pixel Tile Shader | 67 | 55 | -17.9% |
| shader 18 Fur Shader With Anisotropic Lighting | 11 | 10 | -9.1% |
| shader 19 Combination Effect | 37 | 34 | -8.1% |
| shader 22 High Dynamic Range Shader - low quality version without filtering | 31 | 30 | -3.2% |
| shader 23 High Dynamic Range Shader - high quality with fp filtering | 36 | 35 | -2.8% |
We see that there is a fairly mixed set of performance difference between the rendering of these shaders under PS2.0 and PS3.0. In most cases we see a performance degradation, and a reasonably sizable one with Shader 15, but in three cases there is a slight performance improvement.
