Gainward 6800 Ultra Golden Sample

3DMark 2005 score : 4901 (Sharok) and 1957 (Falmir)

Performance summary : Almost fluid

Things are getting better. It's no longer a slow-motion slideshow. 

A review of various cards on a high-end Pentium system can be found at the Tech Report, but lets review my scores first :


Game Tests Sharok Falmir
Return to Proxycon 10.4 6.1
Firefly Forest 6.4 6.2
Canyon Flight 12.5 12.8
Final Score 4901 1957

CPU Tests Sharok Falmir
Return to Proxycon 1.4 1.1
Canyon Flight 2.6 1.8
Final Score 3134 2076

The scores are better than before for Sharok, but nothing stellar. Now that I have splurged on the Pro version of this benchmark, I can give the details and the graphs :

Other Values
Sharok
Falmir
Fill Rate - Single-Texturing (MTexels/s)
3569.8
1958.4
Fill Rate - Multi-Texturing (MTexels/s) 6243.7 3015.1
Pixel Shader (fps)
151.0
48.8
Vertex Shader - Simple (MVertices/s) 46.9
45.6
Vertex Shader - Complex (MVertices/s) 37.0 26.7

This is all a bunch of nice numbers, but what really interests me is the performance given by the graphs. For Proxycon, here is the result :

Sharok

Falmir
Proxycon performance chart
click to enlarge

Proxycon performance graph for Falmir
click to enlarge

The minimum framerate is a dismal 5,95 fps for Sharok, and only 1,36 for Falmir. Dreadful. But with Firefly it gets worse :


Sharok

Falmir
Proxycon performance chart
click to enlarge

Proxycon performance graph for Falmir
click to enlarge

For Sharok, the performance minimum drops by amost 2 frames, Falmir actually gets a 2.4 frame boost. Sure, the 6800 still ups the R9800, but it's by a teeny amount. Oh well, there's still Canyon :


Sharok

Falmir
Proxycon performance chart
click to enlarge

Proxycon performance graph for Falmir
click to enlarge

Here the performance minimum increases across the board, but the end result is still dismal..

As far as performance increase is concerned, Sharok does get a nice boost in this benchmark (4901 over 2474). That's almost double in performance. It could be better, but I suspect that the CPU and chipset are now the limiting factor. Compared to the high-level scores for a quad-pumped P4E running at 3.4Ghz, I am lacking something like 50 points, oh wait. Come to think of it, that difference is laughable.
Maybe I should compare with an AMD 64 4000+, which gives a difference of almost 300 points. Well, that's still not enormous, but it is a respectable difference for this version of 3DMark.


Let's see how the other tests turned out