Page 1 of 1
'make test' hangs vasp 6.4.3 nvhpc_omp_acc
Posted: Wed Jun 18, 2025 2:14 pm
by jaime_gutierrez
Hello,
I've previously successfully compiled vasp 6.4.2 on an HPC cluster with nvhpc_omp_acc as well as nvhpc_omp where the 'make test' have passed.
I am now running into an issue with a new build of version 6.4.3 where the compilation completes, but the 'make test' hangs indefinitely, always at the same spot. I've attached the makefile.include as well as the stdout of the complilation and the 'make test'.
Best Regards,
Jaime
Re: 'make test' hangs vasp 6.4.3 nvhpc_omp_acc
Posted: Fri Jun 27, 2025 1:57 pm
by merzuk.kaltak
Hello,
nvhpc-24.9 + openmpi-3.3.10 is not a validated toolchain.
Can you please choose one from our validated toolchains.
Also, I suggest running with the same number of MPI ranks than GPUs available. This will improve performance and you get rid of following message while running vasp:
Code: Select all
WARNING: INIT_ACC: several MPI-ranks need to share a GPU, which is not
supported by NCCL. The use of NCCL will be switched off. To avoid this,
reduce the number of MPI-ranks: #-of-ranks <= #-of-GPUs (on every node!).
Re: 'make test' hangs vasp 6.4.3 nvhpc_omp_acc
Posted: Mon Jun 30, 2025 5:25 pm
by jaime_gutierrez
Hello Merzuk,
Thank you for your reply.
I do not see where I am using openmpi 3.3.10 in my example. Could you clarify?
Code: Select all
$ which mpiexec
/cm/shared/apps/nvhpc/24.9/Linux_x86_64/24.9/comm_libs/mpi/bin/mpiexec
Code: Select all
$ mpiexec -version
mpiexec (OpenRTE) 4.1.7a1
Report bugs to http://www.open-mpi.org/community/help/
Code: Select all
$ ml list
Currently Loaded Modulefiles:
1) gcc/11.2.0 2) nvhpc/24.9 3) fftw3/openmp/3.3.10 4) hdf5-nvhpc/1.14.0
Based on the above output from my build environment, openmpi 4.1.7 is in use when loading the nvhpc/24.9 module which is also referenced in the makefile.include file.
The references to openmp/3.3.10 for the FFTW related variables in the makefile.include to my understanding are there for referencing the fftw and fftw3_omp libraries which I had compiled specifically for the purposes of a successful Vasp 6.4.2 build and test noted in a previous Vasp forum topic https://www.vasp.at/forum/viewtopic.php?t=20114
Code: Select all
# FFTW (mandatory)
FFTW_ROOT ?= /cm/shared/apps/fftw/openmp/3.3.10
LLIBS += -L$(FFTW_ROOT)/lib -lfftw3 -lfftw3_omp
INCS += -I$(FFTW_ROOT)/include