Dear all,
I am trying to run two psi4 jobs in one machine, both with “-n 8” options. The machine has 16 cores, 1 thread each core (hyper threading is off). However, what I saw from “top” command is that the total “%CPU” of the two jobs is always close to 800%. It seems that the two jobs are sharing the same assigned threads. Did I do anything wrong? What is the correct way to run multiple psi4 jobs in one machine? Thanks.
Michael
Separately executing psi4 -n 8 twice should work and run each psi4 with 8 threads.
quick questions:
- which psi4 version? And how was it installed (conda, self-compiled, etc.)?
- when using top, please type ‘H’ to show the threads (or equivalent command if its a different ‘top’). Does it show 4 or 8 per psi4 instance?
- What kind of calculations are running? (To see if it’s an OpenMP or MKL threading issue)
- I am using Psi4 1.3.2 installed from installer (psi4conda-1.3.2-py37);
- top H does showed that sometimes 8 threads on one psi4 instance. Interestingly, the total %cpu never go beyond 800%. Maybe it is the way %cpu was calculated?
- below is a snapshot that I submited 3 sapt calculations at different time, all with “-n 8” option:
PID PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
8825 20 0 29.985g 0.028t 5248 R 99.9 25.7 91:43.98 psi4
8829 20 0 29.985g 0.028t 5248 R 99.9 25.7 91:51.06 psi4
8830 20 0 29.985g 0.028t 5248 R 99.9 25.7 91:48.89 psi4
8831 20 0 29.985g 0.028t 5248 R 99.9 25.7 91:49.01 psi4
8826 20 0 29.985g 0.028t 5248 R 99.7 25.7 91:51.30 psi4
8827 20 0 29.985g 0.028t 5248 R 99.7 25.7 91:52.25 psi4
8828 20 0 29.985g 0.028t 5248 R 99.7 25.7 91:54.00 psi4
4939 20 0 13.083g 0.011t 1768 R 33.4 10.1 1074:27 psi4
4892 20 0 15.817g 0.014t 2096 R 33.1 12.7 1076:00 psi4
8821 20 0 29.985g 0.028t 5248 R 33.1 25.7 120:29.18 psi4