Published on 11th Aug 2003, written by Dave Baumann for Consumer Graphics - Last updated: 4th Jul 2007
FSAA Performance Comparison
Let's look at the comparative performance of the 5900 Ultra to the 5800 Ultra in the custom SS:SE demo.
4X FSAA
640x480
800x600
1024x768
1280x1024
1600x1200
5800 Ultra
146.5
134.8
103.4
71.4
45.6
5900 Ultra
144.5
140.0
123.4
90.1
66.9
Diff from 5800
640x480
800x600
1024x768
1280x1024
1600x1200
FPS
-2.0
5.2
20.0
18.7
21.3
%
-1%
4%
19%
26%
47%
Here we can really begin to see the difference that the 256-bit bus is making for the rendering performance of the 5900 in relation to the 5800, though this is not all due to the size of the bus width but also the quantity of RAM. Looking at the fill-rate graph we can see that the 5800 Ultra has a slight dip in fill-rate performance at 1600x1200 in comparison to 1280x1024 - this drop off is most likely caused by the 4X framebuffer requirements taking up most of the onboard RAM leaving little room for textures, and some are spilling over to system RAM and will need to be addressed across the AGP bus. The 256MB of RAM on the 5900 Ultra means that there is more available framebuffer space, and hence more room for textures to be stored on the board thus preventing AGP texturing and so no performance is lost by addressing across the comparatively slow AGP bus.
4X FSAA + 8x AF
640x480
800x600
1024x768
1280x1024
1600x1200
5800 Ultra
144.3
114.9
85.9
59.2
39.3
5900 Ultra
144.9
130
99.5
70.5
51.5
Diff from 5800
640x480
800x600
1024x768
1280x1024
1600x1200
FPS
0.6
15.1
13.6
11.3
12.2
%
0%
13%
16%
19%
31%
With both 4x FSAA and 8x AF enabled the performances of the two boards display similar trends, though the performance delta between them is lowered. Because Anisotropic Filtering is more of a fill-rate intensive process this shaves off a little of the bandwidth advantage that the 5900 Ultra has, as does the slight fill-rate advantage the 5800 Ultra has.