[Chimera-users] ATI catalyst 5.7 drivers with R9600SE video card crash winxp when creating surfaces (or running surface benchmark) in chimera-1.2143

Thomas Goddard goddard at cgl.ucsf.edu
Mon Aug 22 11:48:53 PDT 2005

Hi Sabuj,

  I'm surprised how different your Chimera benchmark scores are running
Windows versus Linux on the same machine.

> Chimera-1.2143, WinXP SP1, ATI Mobility Radeon 9600 (128MB), driver 
>, ECS G736 laptop, 1G RAM, P4 3.2GHz
> Linux:   surface 181  mesh 181  contour 146  solid 244  recolor 118
> Windows: surface 350  mesh 150  contour 110  solid 208  recolor 70

The graphics drivers can make a big difference in the Surface / Mesh
scores.  I believe this has to do with the rules the graphics driver
uses to decide whether to put the surface geometry in the graphics
card memory (fast), or shuttle it from the main computer memory for each
drawing (slow).

The difference you see in contouring scores (146 vs 110) is hard to
fathom.  This is not using any graphics -- it is just a calculation done
on the computer CPU.  Do you get close to the same number if you run the
benchmarks twice?  It could be that Windows is much slower at allocating
memory -- that is also done during the contour calculation.

I would definitely *not* add the scores to get an overall measure of
performance.  Some may be much more important to you than others.
Surface and contour are probably most important for looking at density
maps.  Surface is probably most important for looking at large molecular
models.  Recolor is relatively unimportant (it concerns recoloring of
solid volume renderings).

The benchmark code works by increasing the volume data set size until
the drawing or contour calculation speed drops below 10 times per second
(considered "interactive").  Then uses bisection to find exactly what size
volume data set can be handled at 10 frames per second.

See the Chimera User's Guide page for more info about the benchmarks.
It's in the Tools section under Benchmark.


We are currently enhancing the benchmarking tool to time display of
molecular models.


More information about the Chimera-users mailing list