Z Rejection Performance

Dependant on the organisation of the pipelines the Z rejection rate may scale with the shader pipelines. We'll take a look using the GL_REME application that tests various render orders, looking at them in relation to the back to front order, which is the worst case scenario for early pixel rejection.

X1800 XL Overdraw Factor 3 550.3 1791.5 920.3 225.6% 67.2%
Overdraw Facter 8 198.6 1567.3 589.1 689.1% 196.6%
X800 XT Overdraw Factor 3 526.0 1703.8 879.4 223.9% 67.2%
Overdraw Facter 8 190.9 1497.9 565.4 684.6% 196.2%

In general terms there appears to be no difference in the capabilities of the early Z rejection schemes between these two generations of chips at the resolutions tested. ATI state that the Hierarchical Z-Buffer on R520 can reference more pixels, hence higher resolutions are supported, but we are not able to test beyond 1600x1200.




X1800 XL 884.0 602.0 377.0 269.0
X800 XT 695.0 473.0 288.0 207.0

X1800 XL % Faster than X800 XT

27.2% 27.3% 30.9% 30.0%

PowerVR's VillageMark benchmark, which draws a scene with very high overdraw levels, does display a fairly considerable improvement on X1800 XL, but exactly how much of this difference is due to the different vertex rates is difficult to say.

Stencil Performance

Game Test 2 in Futuremark's benchmark application 3DMark03 uses an early Z pass and stencil buffered shadows, so we'll look at the rendering performance of the X1800 XL in this to see how it fares in relation to the X800 XT in a stencil heavy scene.




X1800 XL 157.8 135.6 108.8 83.8 64.9
X800 XT 134.7 115.7 92.4 69.6 53.8

X1800 XL % Faster than X800 XT

17.1% 17.2% 17.7% 20.4% 20.6%

The X1800 XL has a performance advantage over the X800 XT by as much as 21% in this test, but again it is difficult to pinpoint how much of these increases relate to the general increases in architectural efficiency and how much is due to the vertex shader differences.