iOpenShell » Q-Chem specific questions » Is Mem_total per core or per node for parallel calculations?

Is Mem_total per core or per node for parallel calculations?

Page: 1

Author Post
Member
Registered: Nov 2016
Posts: 2
Hello all,
I have what is likely an extremely simple question. If I am running a Q-Chem calculation in parallel over several processors, am I supposed to specify the total memory available to each processor or for the whole node in the Mem_Total field? What about Mem_Static? I tried to do a bit of sleuthing on this forum and in the manual but came up short.

On a somewhat related note, is there any way to get more descriptive error messages on a multi-core calculation? I'll often only get a useful message (for instance if I specify a basis that doesn't support one of my atoms) when the job is run on a single core only. For reference, I'm using Q-Chem 4.0.1.

Thank you very much for your time!
Member
Registered: Dec 2007
Posts: 5
Location: NREL
sgshepard wrote
I have what is likely an extremely simple question. If I am running a Q-Chem calculation in parallel over several processors, am I supposed to specify the total memory available to each processor or for the whole node in the Mem_Total field? What about Mem_Static?


I had the same question--for an MPI-parallelized method, are these numbers per MPI process, or per network-addressable domain (i.e., node)? At the moment, it looks like the former, if I assume mem_static dominates memory usage in TDDFT (top showing 16.5% usage per process, mem_static set to ~15% of node memory).

- Chris
Member
Registered: Dec 2007
Posts: 5
Location: NREL
OK, just ran a node out of memory by doubling # MPI ranks at constant mem_total and mem_static settings, so I'm guessing these are per-rank values.
Member
Registered: Nov 2016
Posts: 2
Thanks for doing that experiment! This is good to know.
Administrator
Registered: Sep 2007
Posts: 175
MEM_STATIC cannot exceed 2 GB (regardless of what you set in the input). It is still used by some routines, but will soon be phased out. If unsure, set it to 2 GB.

Page: 1

iOpenShell » Q-Chem specific questions » Is Mem_total per core or per node for parallel calculations?