We are using preCICE to couple GENE-X simulation with Paraview/Catalyst visualization. In one case, we use preCICE to communicate around 32k points, and we ran into out of memory. We also use slurm with CPU pinning, our aim is to use the idle CPUs on GPU nodes for visualization. I already did this with another library and wanna start a comparison with preCICE. Does anyone have similar experience?
precice-config.xml (1.5 KB)
Hi @Yi_Ju,
I am afraid there is not enough detail to understand what might be wrong here. Do you you have more insights in the meantime?
Overall, your config file looks fine. Visualization:
You are using a nearest-neighbor mapping, which is the cheapest. I assume that every rank of the CPU defines only the mesh that it needs to, and you don’t have duplications.
I also see that you have configured two-level initialization. As precice-cli config check suggests:
To use two-level initialization, both participants need to run in parallel. If you want to run in serial please switch two-level initialization off.
which I assume you already do, but just in case (not sure at which point you run out of memory).