Paraview crashing

39 views
Skip to first unread message

Liam Murray

unread,
Dec 1, 2024, 7:28:42 PM12/1/24
to ProjectChrono

Good afternoon all! hope everyone had a good thanksgiving!
I have been running simulations using DEMdemo_WheelSlopeSlip.cpp and DEMdemo_WheelDP.cpp, followed by post-processing in ParaView. However, when I attempt to visualize all particles in the simulation, ParaView crashes. My system on our HPC has access to over 700 GB of RAM, and I run ParaView on the following hardware:

  • VisRTX 0.1.6 using:
    • Device 0: Tesla V100-PCIE-16GB (Total: 16.9 GB, Available: 13.0 GB)
    • Device 1: Tesla V100-PCIE-16GB (Total: 16.9 GB, Available: 16.2 GB)

For the Wheel Slope Slip simulation, the TableToPoints filter uses about 27.2 GB of RAM. However, when I try the "All Points" visualization option, ParaView crashes every time. I managed to get it working with the "Uniform Spatial Distribution (Bounds Based)" option by limiting the sample points to 1,200,000, which worked fine. That said, I know the simulation contains far more particles than that, as it uses the GRC files we generated.

Is there an alternative setting or workflow I should consider to visualize all the points in the simulation without crashing? Any advice on optimizing memory usage or handling such large datasets in ParaView would be greatly appreciated.

best, 

Liam

Liam Murray

unread,
Dec 1, 2024, 7:29:30 PM12/1/24
to ProjectChrono
i ment to add in when the glyph option is selected for all points 

Ruochun Zhang

unread,
Dec 2, 2024, 8:24:08 PM12/2/24
to ProjectChrono
Hi Liam,

The easy solution would be to reduce the resolution of your glyph, if you are still using the default.

Or you can reduce the number of objects being rendered. For example, use MaskPoints filter to remove the points at the inner parts of the terrain (invisible anyway), then apply glyph.

Or just run simulation with a lower number of particles. If you don't absolutely need this GRC-1 representation, you can give the WheelDPSimplified demo a try. A different 3-sphere clump representation can reduce the overall number of particles to a fraction. However, if you know that eventually your simulation will shoot for millions of particles, then perhaps you shouldn't evade the problem like this.

While I am sure you can tweak Paraview to make it more recommendable for large-scale rendering, like using MPI and OSPRay plugin, I personally just use Paraview for a quick check of the results. For more serious large-scene production, other tools like Blender may be more suitable.

Thank you,
Ruochun

Reply all
Reply to author
Forward
0 new messages