Benchmarks - Overdraw Reduction
Here we'll take a look at the overdraw reduction capabilities of X700's with a couple of tests. The architecture of R420 based products have a portion of Hierarchical Z-Buffer attached to each rendering quad, so we'd expect the overdraw reduction performance to scale evenly with the number of pipelines.
| D3D VillageMark (FPS) | 640x480 | 800x600 | 1024x768 | 1280x1024 | 1600x1200 |
| X700 XT | 391 | 283 | 214 | 156 | 109 |
| X700 PRO | 333 | 304 | 225 | 138 | 97 |
| X800 XT | 855 | 644 | 461 | 275 | 191 |
| X600 XT | 283 | 205 | 134 | 80 | 55 |
| 9800 PRO | 456 | 293 | 180 | 122 | 88 |
| X700 XT % Faster Than: | 640x480 | 800x600 | 1024x768 | 1280x1024 | 1600x1200 |
| X700 PRO | 17.4% | -6.9% | -4.9% | 13.0% | 12.4% |
| X800 XT | -54.3% | -56.1% | -53.6% | -43.3% | -42.9% |
| X600 XT | 38.2% | 38.0% | 59.7% | 95.0% | 98.2% |
| 9800 PRO | -14.3% | -3.4% | 18.9% | 27.9% | 23.9% |
| X700 PRO % Faster Than: | 640x480 | 800x600 | 1024x768 | 1280x1024 | 1600x1200 |
| X800 XT | -61.1% | -52.8% | -51.2% | -49.8% | -49.2% |
| X600 XT | 17.7% | 48.3% | 67.9% | 72.5% | 76.4% |
| 9800 PRO | -27.0% | 3.8% | 25.0% | 13.1% | 10.2% |
The VillageMark tests show a trend that is something close to as we would expect, with the X800 being roughly twice the performance of the X700 XT and then the rest of the boards falling in-line according to their theoretical fill-rate specifications. The X600 XT does fall further behind than its fill-rate differences to the X700 XT and PRO would indicate, however this is because it has no Hierarchical Z-Buffer, and is solely reliant on early Z rejection at the pixel level.
| GL_REME Overdraw Test (FPS) | Factor | Back to Front | Front to Back | Random | % Difference from Back to Front | |
| Front to Back | Random | |||||
| X700 XT | 3 | 455.7 | 2091.9 | 1098.3 | 359.0% | 141.0% |
| 8 | 238.1 | 1825.5 | 703.6 | 666.6% | 195.5% | |
| X700 PRO | 3 | 396.2 | 1785.8 | 936.8 | 350.7% | 136.5% |
| 8 | 201.6 | 1568.4 | 600.4 | 678.0% | 197.8% | |
| X800 XT | 3 | 560.8 | 3763.5 | 2002.3 | 571.1% | 257.1% |
| 8 | 444.9 | 3314.1 | 1298.7 | 644.9% | 191.9% | |
| X600 XT | 3 | 413.7 | 699.4 | 586.5 | 69.1% | 41.8% |
| 8 | 174.5 | 292.9 | 252.0 | 67.8% | 44.4% | |
| 9800 PRO | 3 | 791.0 | 2494.8 | 1312.3 | 215.4% | 65.9% |
| 8 | 286.7 | 2044.4 | 823.6 | 613.1% | 187.3% | |
The GL_REME test is showing some rather curious results in that the Back-to-Front rendering performance with an overdraw factor of 3 as the 9800 PRO is much larger than the other boards, making it appear as though it has less gains from rendering Front-to-Back (which should be optimal for early Z based pixel rejection schemes) or Random ordering. In relation to the relative fill-rate differences the X700’s are also showing an unusually high performance in comparison to the X800 XT with Back-to-Front rendering.
The relative performance differences all appear to stabilise a bit with an overdraw factor of 8, with increases for Front-to-Back and Random ordering within similar ranges for the X700’s, X800 XT and 9800, which does indicate that they have similar Hierarchical Z rejection properties, scaling with the number of pipelines. In both overdraw factor cases, the X600 XT is clearly lagging behind the other due to is lack of Hierarchical Z-Buffer.
