We are trying to compile and run SPECFEM3D global on a Titan machine.
The system has 8 CPU cores and an NVIDIA Titan RTX Graphics Card.
The simulation requires 384 processors. It is the “global_s362ani” example of the SPECFEM3D package. When running with CPU, we get the following error:
" Error wrong number of MPI processes 6 should be 384 , please check…"
I wonder if we can run this example by using the GPU graphic card? If yes, please give me some guide.
a single Titan RTX card with 24GB memory won’t be enough to run the EXAMPLES/global_s362ani_shakemovie/ default example. that example setup requires up to ~120GB memory on GPUs, i.e., you would need multiple Titan cards to run it.
the error you get is a mismatch between the NPROC_** settings in your Par_file (NPROC_XI == NPROC_ETA == 8) and the number of parallel processes you used to run the executable by mpirun -np 6 ... if you want to run it on 6 MPI processes only, you need to set NPROC_XI = 1 and NPROC_ETA = 1 in your Par_file.
for the example to run on your GPU, you would lower the NEX_** settings to around 96 (or 112) which might then just fit into the 24GB limit.