In our recent GeforceFX 5900 Ultra story, we noticed the Nvidia board was well ahead of ATI's Radeon 9800 Pro in Futuremark's 3DMark2003 benchmark, but the two boards were much more evenly matched in our GameGauge application benchmark suite.
We've just discovered certain test anomalies that indicate to us that Nvidia may be special-casing 3DMark2003, throwing away work, to attain higher scores.
We had thought, with the GeForceFX 5900 Ultra now out of the bag, that we were in for a bit of a post-launch lull. But as it turns out, things are just starting to get interesting.
We stated in our GeForceFX 5900 Ultra review that we found the 256MB 5900 Ultra's 3DMark03 results were faster than those delivered by the Radeon 9800 Pro 256MB card, while the 3D GameGauge scores of the two cards were more evenly matched. We found this somewhat odd, and attributed this disconnect to 3DMark03's using vertex and pixel shader code, whereas seven of the eight tests in 3D GameGauge do not.
Our GameGauge suite tests with eight real games, while Futuremark's benchmark uses four synthetic games (Game 1 through 4) that emulate real-world products, but is specifically designed for benchmarking 3D cards. You can read more about Futuremark's benchmark in our detailed analysis from last February.
It turns out, though, that the disparate 3DMark03 scores may have a different root cause: our test results seem to indicate that nVidia is boosting its 3DMark game test scores by cutting corners in the 44.03 DetonatorFX driver we used for testing.
Source:
http://www.extremetech.com/article2/...1086025,00.asp
and Neowin
Hybrid Mode
