Overdraw Reduction

First up let's take a look at the performance of the two boards under the PowerVR VillageMark test.

5800 Ultra 454 326 219 136 96
5900 Ultra 432 310 204 129 92
FPS -22 -16 -15 -7 -4
% -5% -5% -7% -5% -4%

There is a small difference between the two boards with the 5900 Ultra being 4%-7% slower than the 5800 Ultra, which would suggest there is not really any difference between the Z-Cull routines on the 5900.

Now, let's take a look at the effects render order has on GeForce FX to see if we can gauge the effectiveness of its HSR routines. To do this we'll use "Humus"'s GL_EXT_reme benchmark.

Overdraw factor 3 5800 Ultra 648.04 1296.98 911.63
5900 Ultra 673.89 1304.74 935.9
Overdraw factor 8 5800 Ultra 248.17 755.51 504.01
5900 Ultra 258.17 734.09 506.13
Overdraw factor 3 5800 Ultra 100% 41%
5900 Ultra 94% 39%
Overdraw factor 8 5800 Ultra 204% 103%
5900 Ultra 184% 96%

As with VillageMark, the performance of the 5900 under GL_EXT_Reme is behind that of the 5800, which is likely to be a factor of the clock speed difference.


Stencil Performance

To check to see if there are any differences in the stencil rendering performances of 5900 Ultra we'll use another PowerVR test, FableMark

5800 Ultra 158.1 104.7 66.1 40.8 28.7
5900 Ultra 167.4 110.9 69.9 43.0 29.9
FPS 9.3 6.2 3.8 2.2 1.2
% 6% 6% 6% 5% 4%

Here we can see that the 5900 Ultra actually has a small performance advantage over the 5800 Ultra. It's said that the 5800 had issues with Z-Culling whilst stencil rendering and this may be fixed with the 5900 Ultra. Note that this will not be due to UltraShadow since this would require developer support, this can't occur for FableMark since its been around longer than the 5900, and UltraShadow is not exposed through DirectX, which FableMark operates on.