I test the VASP parallel efficiency using same problem, got this bizzare behavior. Can anybody explain why?
Thanks.
CPU# Memory(mb) Time(hrs)
4 4000 13.76
8 4000 6.86
12 4000 10.67
16 4000 7.76
VASP parallel efficiency problem
Moderators: Global Moderator, Moderator
-
fzcsydl
- Newbie
- Posts: 18
- Joined: Tue May 18, 2010 8:54 pm
VASP parallel efficiency problem
Last edited by fzcsydl on Sat Sep 04, 2010 12:04 am, edited 1 time in total.
-
forsdan
- Sr. Member
- Posts: 339
- Joined: Mon Apr 24, 2006 9:07 am
- License Nr.: 173
- Location: Gothenburg, Sweden
VASP parallel efficiency problem
Please understand that there are plenty of factors which effect the calculation times and the scaling. Just to mention a few:
Compiler optimization and architecture specific options, used libraries LAPACK/ScaLAPACK, interconnect between the nodes, used nodes vs. switch configurations, temporary load disruption on the used nodes, available memory/virtual memory and its distribution, VASP settings (e.g. NPAR, LPLANE, NSIM), Number of electronic/ionic steps.
So there is no possibility for us to say anything about why your values look like they do. Please perform systematic investigations for many different settings and monitor the memory usage, and node configurations before you draw any conclusions about the scaling behavior. Also repeat calculations to eliminate the possibility of temporary issues.
Cheers,
/Dan
Compiler optimization and architecture specific options, used libraries LAPACK/ScaLAPACK, interconnect between the nodes, used nodes vs. switch configurations, temporary load disruption on the used nodes, available memory/virtual memory and its distribution, VASP settings (e.g. NPAR, LPLANE, NSIM), Number of electronic/ionic steps.
So there is no possibility for us to say anything about why your values look like they do. Please perform systematic investigations for many different settings and monitor the memory usage, and node configurations before you draw any conclusions about the scaling behavior. Also repeat calculations to eliminate the possibility of temporary issues.
Cheers,
/Dan
Last edited by forsdan on Sat Sep 04, 2010 4:55 pm, edited 1 time in total.