NVIDIA GeForce3 Investigation: What NVIDIA didn't tell us
by Kert Chian on May 15, 2001 1:18 AM EST- Posted in
- GPUs
Anti-aliasing performance
GeForce2 and GeForce3 anti-aliasing involve two stages. First, render samples to framebuffer. Second, blend samples and output anti-aliased image to framebuffer.
GeForce2 vs GeForce3 anti-aliasing
During the first stage of GeForce3 multi-sampling, texture bandwidth requirements are not increased from normal rendering. For GeForce2 supersampling, texture bandwidth is increased from normal rendering by a multiple of the degree of anti-aliasing. The 'super' color pixel buffer bandwidth is similar for both GeForce2 and GeForce3, given the same degree of anti-aliasing. Thanks to Z-compression, 'super' depth buffer bandwidth is lesser for GeForce3. During stage two, GeForce2 and GeForce3 consume the same bandwidth when during sampling for the same degree of anti-aliasing. When writing anti-aliased pixels to the framebuffer, GeForce2 additionally writes to the depth buffer. By virtue of its bandwidth advantage, GeForce3 has a sizeable performance gain over GeForce2 (figure 14).
Figure 14: 32-bit anti-aliasing of GeForce2 and GeForce3
Resolution (32-bit color) | 640x480 | 800x600 | 1024x768 | 1280x1024 | 1600x1200 |
GeForce2, 4-sample AA |
40.9 |
29.2 |
19.5 |
11.1 |
not able |
GeForce2, 3-sample AA |
50.6 |
40.3 |
28.7 |
18.4 |
9.2 |
GeForce3, 4-sample AA |
50.4 |
41.7 |
30.4 |
19.3 |
5.4 |
GeForce3, 2-sample AA |
56.9 |
49.3 |
40.6 |
29.4 |
21.1 |
Quincunx vs 2- and 4-sample GeForce3 multi-sampling
For low resolution anti-aliasing, 4-sample multi-sampling is faster than Quincunx because the latter uses a wider sampling frame, with poor cache coherency. At higher resolutions where bandwidth takes precedence, Quincunx uses less bandwidth than 4-sample multi-sampling. Quincunx only consumes greater bandwidth than 2-sample multi-sampling during the sampling stage two. In addition to retrieving two distinct samples as per 2-sample multi-sampling, Quincunx also retrieves three shared samples.
Figure 15: 16-bit GeForce3 multi-sampling
Resolution (16-bit color) | 640x480 | 800x600 | 1024x768 | 1280x1024 | 1600x1200 |
GeForce3, 2-sample AA |
58.9 |
52.6 |
44.8 |
35.9 |
27.2 |
GeForce3, Quincunx AA |
52.8 |
46.1 |
38.4 |
28.9 |
21.7 |
GeForce3, 4-sample AA |
54.6 |
47.7 |
38.4 |
26.4 |
18.3 |
0 Comments
View All Comments