Originally posted by Thunderbird
Really don't look at all these glxgears scores. Glxgears is not a good benchmark. It mainly draws some polygons to the screen. The results depend on lots of things (driver, cpu(!, yes a very fast gpu can create good scores too), videocard, kernel, resolution, depth ..). If you lets say get 3000 points and someone else gets 6000 points that won't say that ut2004 will run twice as fast on his box. Compare timedemo results of real games and not glxgears scores.
...and timedemo (with individual settings) results are not depending on: driver, cpu, gpu, kernel, color depth,... ;-) You have answered yourself... Yes, glxgears is very basic and not quite good for this.
And the performance is under Linux never the same if: A MDK user with overbloaded creaping KDE 3.x +22 kde-applets sucking the CPU/MEM and 33 daemons sleeping in the bg to suck more. When I came with a pure clean Fedora TWM desktop.
A better thing as bad timedemos and glxgears, because the Unreal engine depends most on the CPU, Q3/ET/Wolfenstein on the GPU maybe... a better way was this spec:
Then we can perhaps make a nice and fair database here, with given attitudes and software and we do not have these threads in future?