[Paraview] parallel rendering problem

Kent Eschenberg eschenbe at psc.edu
Mon May 5 13:53:01 EDT 2008


Is there perhaps one more option: use off-screen rendering? I haven't written OpenGL directly for a while but I think there is a way to use the hardware accelerator to write to a buffer. I'm not sure how to ask PV to use this mode.

I'm particularly interested in the answer because we have a similar system: dual quad-core Xeons plus a graphics card on each node, with 4 nodes.

Kent
Pittsburgh Supercomputing Center

Berk Geveci wrote:
> The problem is that the server windows are overlapping. Most OpenGL
> drivers, specially on Linux, does not fully render in regions that are
> obscured by other windows. There are a few ways to fix this:
> 0. Make sure the windows do not overlap
> 1. Do not use parallel rendering - disable remote rendering in your settings
> 2. Use software rendering by compiling against Mesa
> 
> Which one is the best depends on what you are trying to do. It sounds
> like you are running the server on the same machine as the client. In
> that case, (1) may be the better option as long as you are not volume
> rendering. If you are running on a cluster, (0) is the best option.
> You can achieve that by setting the DISPLAY properly to make sure that
> each window is rendering on a separate display. (2) is the fallback
> option but is probably the slowest unless you have a lot of nodes.
> 
> -berk
> 
> 
> On Mon, May 5, 2008 at 1:00 PM, Server Levent Yilmaz
> <leventyilmaz at gmail.com> wrote:
>> Hi,
>>
>> I have what is seemingly a parallel rendering problem, although I can not be
>> sure.  I have tested this with ParaView 3.2.1 and the latest HEAD (as of
>> this email's date). Please take a look at these screenshots: sequential (no
>> problems here), 4 CPU run with 3.2.1, 4 CPU run with HEAD.
>>
>> The system is,
>> cpu: dual quad core Intel Xeon
>> gpu: dual Nvidia Quadro FX 570
>> os: Ubuntu 8.04 (hardy)
>> libs: gcc -4.2.3, mpich2-1.0.7 (with ch3:nemesis)
>>
>> I'd greatly appreciate any clues to track down the problem.
>>
>> thanks
>> L:event
>>
>> --
>> Server Levent Yilmaz
>> Mechanical Engineering
>> University of Pittsburgh


More information about the ParaView mailing list