I can't seem to be able to run a parallel version of VASP either for exact diagonalization, or for the ACFDT-RPA algo. In both cases, once the WAVECAR is loaded, I get a segmentation fault. This is not a memory issue, as everything runs smoothly using a single MPI process.
I have compiled VASP using intel compilers and MKL from 2018 (v4). The cluster on which I am running uses intel's Omni-path architecture for inter-node communication. I use the following settings to run VASP:
Code: Select all
export I_MPI_FABRICS=shm:ofi export I_MPI_OFI_PROVIDER=psm2