Vertex Shader Performance

D3D RightMark was used to assess how well the boards coped with high levels of geometry (along with diffuse & specular lighting) to process, via a range of shader models.

7800 GTX 512MB 87.0 76.9 77.3 49.3 60.4 60.4
7800 GTX 256MB 76.2 66.7 67.3 41.9 51.8 52.0
6800 Ultra 256MB 57.3 47.9 48.1 29.8 37.0 37.2
7800 GTX 256MB 14.2% 15.3% 14.8% 17.6% 16.6% 16.1%
6800 Ultra 256MB 51.8% 60.4% 60.8% 65.8% 63.1% 62.5%

The vertex shader performance difference between the GTX 512MB and the GTX 256MB is in the mid-to-high teens, which is what we would expect given the theoretical rates, and that the geometry rates are the smallest differentiators between the two parts.

Pixel Shader Performance

G70 has already proven itself to be have a very powerful fragment pipeline. Here we'll see how the increased clockspeeds of the 7800 GTX 512MB increases the Pixel Shader performance.

PS1.1 Psychodelic procedural 883.9 697.2 414.8 26.8% 113.1%
PS1.4 Wood procedural 727.5 583.3 303.0 24.7% 140.1%
PS2.0 Marble procedural 464.8 359.9 190.7 29.1% 143.7%
PS2.0 Blinn 1 Light (FP) 444.1 350.7 191.3 26.7% 132.2%
PS2.0 Blinn 1 Light (PP) 480.1 361.2 240.1 32.9% 100.0%
PS2.0 Phong 3 Lights (FP) 241.2 195.7 95.6 23.3% 152.5%
PS2.0 Phong 3 Lights (PP) 279.8 219.8 120.0 27.3% 133.1%
PS2.x Cook-Torrance 3 Lights (FP) 87.6 68.5 39.8 27.9% 119.8%
PS2.x Cook-Torrance 3 Lights (PP) 118.0 93.5 59.7 26.3% 97.6%

With these pixel shader tests the performance difference between the GTX 512MB and the GTX 256MB are in the region of theoretical differences. With the changes to the G70 fragment pipeline in relation to NV40, the increases in fragment and texture pipelines and the larger clock speeds, the 7800 GTX 512MB is nearly always over twice as fast as the 6800 Ultra in these tests.