On Thu, 2008-06-05 at 20:41 +0200, Moreno wrote:

> To make comparisons between different machines glxgears is certainly
> inappropriate but to make comparisons on the same machine under
> constant conditions (same resolution, same color depth, same running
> program etc.) this program is more than enough.

No, it's not. It's testing an insignificant operation in a completely
atypical context (i.e. you never actually *do* anything with your card
which involves simple, non-texture mapped polygons doing the same
operation over and over and over again in completely empty space).

> If starting glxgeras with a kernel the wheels turn fluids and with an
> other kernel turn stepwise i do not need other test.

Yes, you do; quite apart from the above, glxgears suffers from the
classic rotating wheel problem, where it can look perfectly fluid and
smooth at 1 zillion FPS and look like it's incredibly jagged and
spinning backwards at 1 zillion minus 1 FPS. The numbers you reported
are not anywhere near a level at which it would *actually* be
'jerking' (which would be less than 15fps).

> However, if you suggest a more accurate test I'm willing to repeat the analysis.

Any real-world application - run a timedemo in Quake3, or something.
--
adamw