Page 1 of 1

Significant Time Consumption Issue during a specific VASP Calculation.

Posted: Sat Jan 13, 2024 12:47 pm
by hszhao.cn@gmail.com
Dear VASP Forum,

I am writing to seek some guidance regarding the issue I recently encountered while using the VASP.6.4.2 for a calculation given in the attachment. I am experiencing significant time consumption issues related to this calculation, and I would like to know whether this is reasonable in this case.

This task has been done using 12 cores on my X10DAi server with two E5-2699V4 processors running Linux, and upon completion of the task, the time performance is as follows:

Code: Select all

$ tail -15 OUTCAR
  
 General timing and accounting informations for this job:
 ========================================================
  
                  Total CPU time used (sec):    29973.172
                            User time (sec):    29316.019
                          System time (sec):      657.152
                         Elapsed time (sec):    29983.274
  
                   Maximum memory used (kb):      587732.
                   Average memory used (kb):          N/A
  
                          Minor page faults:     32556825
                          Major page faults:          617
                 Voluntary context switches:         1993
I truly appreciate your assistance and hope to find some possible solutions or suggestions so that I can optimize for efficiency as much as possible.

Kind regards,
Zhao

Re: Significant Time Consumption Issue during a specific VASP Calculation.

Posted: Sat Jan 13, 2024 12:51 pm
by hszhao.cn@gmail.com
I forgot to attach related input and output files, here it is.

Regards,
Zhao

Re: Significant Time Consumption Issue during a specific VASP Calculation.

Posted: Mon Jan 15, 2024 8:10 am
by alex
Hi Zhao,

you may consider a couple of things:
- you are doing a semiconductor or insulator, hence is a tight k-mesh necessary (or maybe just for the final iteration in an extra simulation for some kind of sanity check)?
- you are doing many structure optimisation steps: hard to compare to anything! The single steps look OK.
- Gamma centering is good. Sometimes uneven numbers are even better.
- with a small machine like yours it's always good to save a bit here and there (if you respect the physics).

Best regards,

alex