Games Benchmarks - DirectX
UT2004
The first game benchmark we'll look at is UT2004. To our knowledge UT2004 is still largely based on a similar build of engine as UT2003, which means its feature set utilisation still marks it primarily as a DirectX7/8 engined title. We are using a custom Firing Squad benchmark demo.
| UT2004 (FPS) | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
| S18 Nitro | 71.0 | 69.5 | 54.7 | 39.4 | 28.7 |
| X600 PRO | 69.8 | 64.4 | 51.1 | 37.5 | 26.6 |
| 6600 | 81.8 | 79.4 | 69.4 | 52.9 | 37.9 |
| 6200 | 79.2 | 68.6 | 50.1 | 36.3 | 25.8 |
| S18 % Faster Than: | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
| X600 PRO | 1.7% | 7.8% | 7.1% | 5.2% | 8.2% |
| 6600 | -13.2% | -12.5% | -21.2% | -25.6% | -24.2% |
| 6200 | -10.4% | 1.2% | 9.2% | 8.5% | 11.3% |
With normal rendering under our UT2004 test we see that the S18 Nitro is able to outperform the X600 PRO and 6200, but is behind the 6600 by about 25% at high resolution.
| UT2004, 2x FSAA & 4x AF (FPS) | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
| S18 Nitro | 42.1 | 32.3 | 22.8 | 16.2 | 11.6 |
| X600 PRO | 56.8 | 45.4 | 34.7 | 23.7 | 15.3 |
| 6600 | 76.2 | 65.5 | 49.0 | 36.4 | 25.7 |
| 6200 | 62.7 | 48.0 | 34.2 | 24.7 | 16.8 |
| S18 % Faster Than: | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
| X600 PRO | -25.8% | -29.0% | -34.5% | -31.8% | -24.0% |
| 6600 | -44.8% | -50.8% | -53.5% | -55.5% | -54.9% |
| 6200 | -32.9% | -32.8% | -33.5% | -34.5% | -30.9% |
We can see here that even with just 2x FSAA and 4x Anisotropic Filtering (AF) enabled the performance hit for SuperSampling FSAA on the S18 is much greater than that of the MultiSample schemes on the other boards as the S18 now has a performance disadvantage to them all.
| S18 Nitro (FPS) | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
| Normal | 71.0 | 69.5 | 54.7 | 39.4 | 28.7 |
| 4x AF | 69.6 | 57.5 | 42.0 | 30.7 | 22.5 |
| 2x FSAA | 60.8 | 45.6 | 32.1 | 22.2 | 15.4 |
| 2x FSAA & 4x AF | 42.1 | 32.3 | 22.8 | 16.2 | 11.6 |
| % Diff from Normal | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
| 4x AF | -2.0% | -17.3% | -23.3% | -22.2% | -21.7% |
| 2x FSAA | -14.3% | -34.4% | -41.2% | -43.6% | -46.5% |
| 2x FSAA & 4x AF | -40.7% | -53.6% | -58.4% | -58.9% | -59.6% |
Looking at the individual performances of each of the IQ options enabled independently on the S18 we see that enabling 4x AF accounts for about a 23% performance hit at worst, even though only Bilinear is utilised when Anisotropic Filtering is enabled. Utilising 2x FSAA on the other hand, is approaching half the normal rendering performance at high resolution, however this is consistent with SuperSampling as all the work for every sub-sample doubles the rendering requirements with 2x FSAA.
