I am once again building and running NWChem following an 8-year hiatus.
I have built NWChem 6.6 using Intel 17.0.4 compilers and OpenMPI (from Mellanox HPC-X). I have successfully run siosi8.nw on 64 and 96 nodes (Broadwell 14-cores/socket, 2 cores/node), but running it on 48 nodes or fewer causes the program to crash with "dft_3cincor: blew out of dbls 999".
On 48 nodes, the following are the last three lines before the error:
54.537 MW buffer allocated for incore 3-center
2e- integral storage on stack.
nlast n3cdbl 54783320 54536525
I read the 'Error message in "dft 3cincor"' thread from Nov. 2016, especially Edo Apra's replies. Adding 'set dft:n3c_dbl' with various large values up to and including 1024000000 have failed to help.
What am I missing?
Saludos,
Gerardo
--
Gerardo Cisneros-Stoianowski, Ph.D.
HPC Applications Performance Engineer
Mellanox Technologies
|