How to choose one GPU in dual GPU system? (SOLVED)

[yoyokalf5]

We have a dual GPU server with Tesla P40 and M40, which have different architectures. I must turn off manually in one GPU to run SPECFEM3D. How to instruct the SPECFEM3D to run using GPU 0 or 1 without turning off one GPU?

[tianshi-liu]

Is "CUDA_VISIBLE_DEVICES=0 mpirun -np $NPROC ./bin/xspecfem3D" what you want? It tells the system to use GPU 0 only, and you can use "nvidia-smi" to check which GPU is 0 or 1. If you don't use MPI then "CUDA_VISIBLE_DEVICES=0 ./bin/xspecfem3D" should be enough.

[yoyokalf5]

Thank you so much, It's work.