diff serial/parallel versions ?

Message

pmignon · #1 Post by **pmignon** » Mon Feb 26, 2007 10:02 am

Hello,

I am running vasp on a linux cluster. the serial and parallel versions have benn installed on this cluster. To check the validity of the parallel version i run the same calculation using the serial and parallel versions (the INCAR? POSCAR POTCAR and KPOINTS are the same). The two calculations give different output.

Serial:
1 F= -.12106489E+04 E0= -.12105946E+04 d E =-.121065E+04
2 F= -.11677569E+04 E0= -.11677569E+04 d E =0.428920E+02
3 F= -.12120875E+04 E0= -.12120390E+04 d E =-.143862E+01
4 F= -.12125392E+04 E0= -.12124938E+04 d E =-.189029E+01
5 F= -.12128380E+04 E0= -.12127974E+04 d E =-.218904E+01
6 F= -.12107124E+04 E0= -.12106630E+04 d E =0.212559E+01
7 F= -.12135512E+04 E0= -.12135076E+04 d E =-.713236E+00
8 F= -.12125947E+04 E0= -.12125636E+04 d E =0.956468E+00
9 F= -.12140110E+04 E0= -.12139736E+04 d E =-.459860E+00
10 F= -.12140975E+04 E0= -.12140584E+04 d E =-.864838E-01

Parallel:
1 F= -.12106489E+04 E0= -.12105946E+04 d E =-.121065E+04
2 F= -.11678091E+04 E0= -.11678091E+04 d E =0.428398E+02
3 F= -.12120882E+04 E0= -.12120399E+04 d E =-.143933E+01
4 F= -.12125402E+04 E0= -.12124948E+04 d E =-.189124E+01
5 F= -.12128377E+04 E0= -.12127971E+04 d E =-.218875E+01
6 F= -.12107349E+04 E0= -.12106863E+04 d E =0.210276E+01
7 F= -.12135514E+04 E0= -.12135078E+04 d E =-.713757E+00
8 F= -.12125822E+04 E0= -.12125514E+04 d E =0.969220E+00
9 F= -.12140110E+04 E0= -.12139734E+04 d E =-.459616E+00
10 F= -.12140978E+04 E0= -.12140587E+04 d E =-.867461E-01

It is normal that I get these differences

thank you in advance.
PIerre.

pmignon · #2 Post by **pmignon** » Mon Feb 26, 2007 12:34 pm

Hello,

I checked again the parallel and serial versions for the bench.Hg and benchmark examples. I realized that the only differences are for some electronic iteration, but the ionic steps for benchmark are the same. However, my OSZICAR is really different than the OSZICAR.ref and OSZICAR.ref.pc given with the examples.
How to check my OSZICAR

thank you,
pierre.

#3 Post by **admin** » Mon Mar 05, 2007 3:50 pm

if you compare the results of serial and parallel runs, please check if the number of bands was the same in both runs (NBANDS always is an integer multiple of the number of nodes). Please set NBANDS of the serial run explicitely to the value of the parallel run. (please have a look at the handbook, chapter Theoretical background, algorithms for further explanation)