Benchmarks - Pixel Shader (RightMark3D)

We'll take a further look at the shader performance of the Radeon X700's using some pixel shader tests from Rightmark3D.

 

Procedural Wood (PS1.4) 192.7 155.5 370.3 91.7 152.1
Procedural Marble (PS2.0) 246.6 224.3 540.9 152.6 219.9
Procedural Marble - PS2.0 FP16 256.1 239.9 559.9 141.7 196.5
Lighting (Blinn) - PS2.0 229.8 200.1 463.3 125.7 187.4
Lighting (Blinn) - PS2.0 FP16 230.4 200.5 464.9 121.6 191.7
Lighting (Phong) - PS2.0 127.8 114.0 265.0 68.7 103.9
Lighting (Phong) - PS2.0 FP16 128.0 115.2 266.4 68.2 103.8
 
Procedural Wood (PS1.4) 24.0% -48.0% 110.3% 26.8%
Procedural Marble (PS2.0) 9.9% -54.4% 61.6% 12.2%
Procedural Marble - PS2.0 FP16 6.8% -54.3% 80.8% 30.4%
Lighting (Blinn) - PS2.0 14.9% -50.4% 82.8% 22.6%
Lighting (Blinn) - PS2.0 FP16 14.9% -50.4% 89.5% 20.2%
Lighting (Phong) - PS2.0 12.0% -51.8% 86.0% 22.9%
Lighting (Phong) - PS2.0 FP16 11.0% -52.0% 87.6% 23.2%
 
Procedural Wood (PS1.4) -58.0% 69.6% 2.2%
Procedural Marble (PS2.0) -58.5% 47.1% 2.0%
Procedural Marble - PS2.0 FP16 -57.1% 69.4% 22.1%
Lighting (Blinn) - PS2.0 -56.8% 59.2% 6.7%
Lighting (Blinn) - PS2.0 FP16 -56.9% 64.9% 4.6%
Lighting (Phong) - PS2.0 -57.0% 66.0% 9.7%
Lighting (Phong) - PS2.0 FP16 -56.7% 69.0% 11.0%

In this case the relative performances differences in the first few tests show similar trends as with the 3DMark tests. The performance difference between the final four tests, however, fall much more inline with the theoretical fill-rate differences between the boards, indicating that they probably have longer instruction length, making them less reliant of texture performance.

RightMark Vertex Shader Tests

We'll use RightMark3D again to look at the geometry performances of the Radeon X700's.

 

X700 XT 44.7 42.6 42.9 42.8
X700 PRO 39.6 37.6 38.0 38.1
X800 XT 46.9 44.3 44.9 44.9
X600 XT 16.7 15.1 15.4 15.4
9800 PRO 25.1 22.6 23.0 23.0
 
X700 PRO 12.8% 13.0% 12.8% 12.4%
X800 XT -4.7% -4.0% -4.5% -4.7%
X600 XT 167.8% 181.6% 178.0% 177.4%
9800 PRO 78.5% 88.0% 86.3% 86.2%
 
X800 XT -15.5% -15.1% -15.3% -15.2%
X600 XT 137.4% 149.1% 146.5% 146.8%
9800 PRO 58.2% 66.3% 65.2% 65.6%

Pure geometry processing isn’t quite as reliant on other influences such as bandwidth (unless there are plenty of other operations going on that constraining the bandwidth – which probably isn’t the case in these tests) hence the performance differences between all the boards follow very closely to their geometry performance differences. This test illustrates how close in geometry performance the X700’s are to the higher end X800 XT – not many games at this point in time are largely geometry limited, although a few may be to some extent, so the gains from vertex processing may not be too frequent with these types of applications, however should ATI move this chip into the workstation market, its geometry performance relative to costs should be favourable for it.