http://kegel.com/wine/yagmarkdata/ now has data from 3dmark2000, 2001, and 06 running on a beefy i7 960 with an equally beefy Nvidia GTX 295 with either Windows 7 or Linux+Wine (all tests run at 1024x768).
Here are the results of one run for each plucked at random (sorry, no breakdown of 3dmark06 results yet):
Win 7 Wine units 3dmark2000_3DMark_Result: 39587 26278 3Dmarks 3dmark2000_CPU_Speed: 1269 1276 CPU_3Dmarks 3dmark2000_Game_1_Helicopter_Low_Detail: 689.7 415.2 FPS 3dmark2000_Game_1_Helicopter_Medium_Detail: 526.8 319.4 FPS 3dmark2000_Game_1_Helicopter_High_Detail: 412.1 260.1 FPS 3dmark2000_Game_2_Adventure_Low_Detail: 981.2 648.4 FPS 3dmark2000_Game_2_Adventure_Medium_Detail: 429.0 334.9 FPS 3dmark2000_Game_2_Adventure_High_Detail: 260.2 211.8 FPS
3dmark2001_3DMark_Score 46297 28194 3DMarks 3dmark2001_Game_1_Car_Chase_Low_Detail 551.3 314.6 fps 3dmark2001_Game_1_Car_Chase_High_Detail 111.5 133.0 fps 3dmark2001_Game_2_Dragothic_Low_Detail 678.0 462.2 fps 3dmark2001_Game_2_Dragothic_High_Detail 399.6 184.2 fps 3dmark2001_Game_3_Lobby_Low_Detail 602.7 371.5 fps 3dmark2001_Game_3_Lobby_High_Detail 294.0 135.5 fps 3dmark2001_Game_4_Nature 593.9 382.9 fps
3DMark06_result: 21519 6038 3dmarks
This might not be all that useful on its own, but once I have daily automatic runs going, it ought to be helpful in looking for performance regressions.
These results are probably only accurate to +- 5%. Later on I'll do multiple runs and compute the median to reduce the noise level.
Notes:
On really big screens, at least under wine, 3dmark2000 seems to lose its mind and put up a warning dialog behind the main window (thus hanging the test). I had to set the screen size down to something closer to 1024x768 for it to calm down and run the benchmark.
3dmark2000 and 3dmark2001 need to be run in xp compatibility mode on Windows 7, or they complain of insufficient memory.
Again, this was done by running http://code.google.com/p/winezeug/source/browse/trunk/yagmark - Dan