Texture Filtering Performance Comparisons
Using the same setting in SS:SE as previously, we'll do a few more comparisons of the texture filtering performance.

8X | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
Aggressive | 138.2 | 135.8 | 133.7 | 119.7 | 93.4 |
Balanced | 139.5 | 137.5 | 128.9 | 99.2 | 72.3 |
Application | 138.0 | 132.8 | 109.4 | 77.2 | 55.3 |
% Diff From Aggressive | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
Balanced | 1% | 1% | -4% | -17% | -23% |
Application | 0% | -2% | -18% | -36% | -41% |
Here we are comparing the highest filtering level of the three quality level setting available on GeForce FX 5800. As we can see they give quite a wide range of performances with the Balanced mode being, at worst, 23% slower than the Aggressive mode, and the Application mode being 41% slower.

640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 | |
Bilinear | 139.1 | 137.2 | 134.6 | 122.1 | 96.8 |
Trilinear | 138.9 | 134.6 | 112.4 | 78.0 | 55.1 |
% Diff | 0% | -2% | -16% | -36% | -43% |
Taking a look at the performance of plain Tri or Bilinear filtering under the 'Application' mode we can see that there is quite a large performance difference between the two modes. This would indicate that each of the texture units are bilinear capable and require two units to perform Trilinear Filtering.

8X AF ('Application') | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
GeForce 4 Ti4600 | 118.6 | 86.4 | 56.4 | 35.3 | 24.3 |
GeForce FX | 138.0 | 132.8 | 109.4 | 77.2 | 55.3 |
GeForce FX (400/400) | 137.7 | 125.6 | 95.1 | 65.5 | 46.8 |
% Diff from Ti4600 | 640x480 | 800x600 | 1024x768 | 1280x960 | 1600x1200 |
GeForce FX | 16% | 54% | 94% | 119% | 128% |
GeForce FX (400/400) | 16% | 45% | 69% | 86% | 93% |
Comparing the GeForce FX's and GeForce4's 8x Anisotropic Filtering in 'Application' mode, which is the closest to GeForce4's filtering quality, we can see that the 5800 has quite a performance advantage over GeForce4. The highest performance difference is 128%, which is beyond the straight fill-rate differences of the two boards.