Fill rate
Theoretical fill rate and memory bandwidth peaks don't necessarily dictate real-world performance, but they're a good place to start.
| Core clock (MHz) | Pixel pipelines | Peak fill rate (Mpixels/s) | Texture units per pixel pipeline | Peak fill rate (Mtexels/s) | Memory clock (MHz) | Memory bus width (bits) | Peak memory bandwidth (GB/s) | |
| Radeon 9600 Pro | 400 | 4 | 1600 | 1 | 1600 | 600 | 128 | 9.6 |
| Radeon 9600 XT | 500 | 4 | 2000 | 1 | 2000 | 600 | 128 | 9.6 |
| GeForce FX 5600 Ultra | 400 | 4 | 1600 | 1 | 1600 | 800 | 128 | 12.8 |
| GeForce FX 5700 Ultra | 475 | 4 | 1900 | 1 | 1900 | 906 | 128 | 14.4 |
| GeForce FX 5900 XT | 400 | 4 | 1600 | 2 | 3200 | 700 | 256 | 22.4 |
| GeForce FX 5900 | 450 | 4 | 1800 | 2 | 3600 | 850 | 256 | 27.2 |
With a core clock speed of 400MHz, the GeForce FX 5900 XT can't quite match the 5700 Ultra or Radeon 9600 XT's single texture fill rate. However, the 5900 XT's multi-texturing fill rate blows away the competition, as does its peak memory bandwidth, which more than doubles what's available with the Radeon 9600 XT. It's also worth noting that the 5900 XT's NV35 graphics chip can, under certain circumstances, look a lot like an eight-pipe chip.
Of course, theoretical peaks don't always hold up in the real world. How does the 5900 XT deliver on its fill rate potential?
Despite having a lower theoretical single-texturing fill rate peak than both the 5700 Ultra and Radeon 9600 XT, the GeForce FX 5900 XT blows them both away in 3DMark03's single-texturing fill rate tests. The 5900 XT is way out ahead in our multi-texturing test, too, nearly doubling the performance of the 5700 Ultra.
Shader performance
Given the GeForce FX's unconventional shader architecture, it's not easy to come up with theoretical expectations for pixel or vertex shader performance, so let's cut right to the graphs.
The GeForce FX 5900 XT leads the pack in both pixel and vertex shader performance, though pixel shader performance drops noticeably with the latest drivers. Those same drivers dramatically increase vertex shader performance. Win some, lose some, I guess.
