[Paraview] Apparent problems with NVIDIA driver on Linux

Cory Quammen cquammen at cs.unc.edu
Tue Feb 21 13:51:50 EST 2012


I wanted to post some information to hopefully save others time in
case they are running into rendering oddities while saving an
animation.

Summary

We have seen rendering problems while generating animations from
ParaView 3.12.0 64-bit using the NVIDIA 295.20 driver on both Red Hat
Enterprise Linux Workstation release 6.1 and Scientific Linux release
6.1 (Carbon). Both machines on which we observed the problem have
NVIDIA Quadro 5000 cards. Switching to NVIDIA driver version 275.43 on
both operating systems seems to have resolved our rendering issues.
This driver is hard to find from the main NVIDIA driver web site, but
you can get it here:

http://www.nvidia.com/object/linux-display-amd64-275.43-driver.html

Details

We were trying to generate a simple animation of a time series of
point particles visualized as glyphs. In several of the frames, a
subset of the spheres -- all produced from one glyph filter -- would
appear to switch to the Point representation from the Surface
representation we wanted. This would only happen for a few
non-consecutive frames in a 300 frame animation. They were always the
same frames, so I initially suspected some change in ParaView or VTK
was leading to this behavior.

After compiling and testing an embarrassing number of different
commits in ParaView's git history, I narrowed it down to the
v3.12.0-RC1 tag. The rendering problem didn't appear for any revision
prior to the v3.12.0-RC1 tag, but it was present at the tag and every
commit after it. After looking at the source code, there didn't appear
to be a change to any relevant rendering code in ParaView or VTK, much
less anything that could have caused this problem.

Giving up on the software side, I decided to try a different NVIDIA
driver. I had version 295.20 installed when I observed the rendering
problems. Installing version 275.43 (a more recent driver release, but
lower driver number, for whatever reason) seems to have resolved the
problem for ParaView 3.12.0.

I don't have an explanation for why the ParaView revision seems to
have been involved, other than perhaps the size of the binary somehow
led to the error in the NVIDIA driver.

Here's hoping I save someone some time,
Cory

-- 
Cory Quammen
Research Associate
Department of Computer Science
The University of North Carolina at Chapel Hill


More information about the ParaView mailing list