Skip to first unread message
mattasplund
Sep 16, 2020, 4:20:09 PM9/16/20
to NWChem Forum
I am trying to get nwchem 7.0 running on our supercomputer cluster, but having real problems. I have compiled 7.0 with gcc 9, openmpi4.0.2 and openblas and 6.8 with gcc 8, openmpi 3.1 and openblas. Both run just fine with <4 processors, but with more than that they crash with a sigsegv error, but not immediately, often getting through a number of iterations before the crash. The backtrace puts the error in ga-5.7.1/comex/src-mpi/comex.c
I have worked with our supercomputer folk, but cannot see where the problem is. Any suggestions about how to attack this?
Matthew Asplund
Department of Chemistry and Biochemistry
Brigham Young University
Provo, UT
(801) 422-5275
Department of Chemistry and Biochemistry
Brigham Young University
Provo, UT
(801) 422-5275
Edoardo Aprà
Sep 16, 2020, 4:24:43 PM9/16/20
to NWChem Forum
My recommendation is to recompile the tools with ARMCI_NETWORK=MPI-PR
The default ARMCI_NETWORK=MPI-TS should only be used when there is a moderate degree of parallelization.
See also
mattasplund
Sep 16, 2020, 7:41:39 PM9/16/20
to NWChem Forum
Perfect. That fixed it.
Reply all
Reply to author
Forward
0 new messages