NWchem with Gotoblas2 but poor performance


Click here for full thread
Clicked A Few Times
Quote:Tpirojsi Oct 8th 6:04 pm
Quote:Edoapra Oct 8th 4:48 pm
Have you set the env. variables

OMP_NUM_THREADS

and

GOTO_NUM_THREADS

to 1?
http://www.tacc.utexas.edu/tacc-projects/gotoblas2/faq



Edo, I didn't do anything about this 2 env variables. Do they need to be set in the submit script, in compiling step, or in .bash_profile? Also, the no. of thread to be set needs to be the same as the no. of cores used to run a job, right?

Thank you,
Tee


Edo, never mind above questions. I set one of those environments (GOTO_NUM_THREADS) to 1 in submit script and tried running a job. It seemed to work perfectly now.

However, The time used to perform a test case for 12 and 24 cpus (or up to 36 cpus) are approximately the same. This might be because of the small size of problem?