[CP2K:10729] Re: How to accelerate the calculation of pdos ?

Krack Matthias (PSI) matthia... at psi.ch
Fri Sep 14 09:27:18 CEST 2018


Hi

As reported in the CP2K output headers

GLOBAL| Total number of message passing processes                             1
GLOBAL| Number of threads for this process                                   36

both runs were not MPI parallel and only OpenMP parallel using the same resources, i.e. one MPI process and 36 OpenMP threads. Thus it is not surprising that you observed no acceleration, since you used the same resources. Moreover, the use of more than 8 threads per (MPI) process is rarely beneficial for CP2K runs and will result rather in a slowdown than a speedup. You have to launch the runs properly, e.g. using something like “mpiexec –n 12 cp2k.popt” which depends, of course, on your installation. If you are using a cp2k.psmp executable, then you should set in addition “export OMP_NUM_THREADS=1”.

Matthias

From: cp... at googlegroups.com <cp... at googlegroups.com> On Behalf Of Tianshu Jiang in Beijing
Sent: Freitag, 14. September 2018 03:50
To: cp2k <cp... at googlegroups.com>
Subject: [CP2K:10729] Re: How to accelerate the calculation of pdos ?

Hi Krack, thanks for your reply.
In the attachment, bilayerIso.out is the file output by 1 core and bilayerIsoPara.out is the file output by 12cores.
I have no idea where the problem is.

在 2018年9月12日星期三 UTC+8下午7:58:50,Matthias Krack写道:
Hi Tianshu Jiang

without providing the CP2K output files of your 1-core and 12-cores runs, it is quite unlikely that you will get any reasonable hint from someone in this forum.

Matthias

On Wednesday, 12 September 2018 04:41:07 UTC+2, Tianshu Jiang in Beijing wrote:
Hi everyone in cp2k community,

I am using cp2k to calculate the pdos of graphene, but the time spent to completing the calculation when I use 1 core and 12 cores is the same.
The version I compile is Linux-x86-64-gfortran, and I use cp2k.psmp to finish the job.
But from the *.out file I get that in both situation (1 core and 12 cores), the job finished after half an hour from beginning.
My question is how can I accelerate the calculation using parallel computing ?

The following is my inputfile. Thanks for your reply !
&GLOBAL
  PROJECT trilayerABCIso
  RUN_TYPE ENERGY
  PRINT_LEVEL MEDIUM
&END GLOBAL

&FORCE_EVAL
  METHOD Quickstep
  &DFT
    BASIS_SET_FILE_NAME  BASIS_MOLOPT
    POTENTIAL_FILE_NAME  POTENTIAL

    &POISSON
      PERIODIC XYZ
    &END POISSON
    &SCF
      SCF_GUESS ATOMIC
      EPS_SCF 1.0E-6
      MAX_SCF 300

      # The following settings help with convergence:
      ADDED_MOS 100
      CHOLESKY INVERSE
      &SMEAR ON
        METHOD FERMI_DIRAC
        ELECTRONIC_TEMPERATURE [K] 300
      &END SMEAR
      &DIAGONALIZATION
        ALGORITHM STANDARD
        EPS_ADAPT 0.01
      &END DIAGONALIZATION
      &MIXING
        METHOD BROYDEN_MIXING
        ALPHA 0.2
        BETA 1.5
        NBROYDEN 8
      &END MIXING
    &END SCF
    &XC
      &XC_FUNCTIONAL PBE
      &END XC_FUNCTIONAL
    &END XC
    &PRINT
      &PDOS
        # print all projected DOS available:
        NLUMO -1
        # split the density by quantum number:
        COMPONENTS
      &END
      &E_DENSITY_CUBE ON
          STRIDE 1 1 1
      &END E_DENSITY_CUBE
    &END PRINT
  &END DFT

  &SUBSYS
    &CELL
      # create a hexagonal unit cell:
      ABC  [angstrom] 2.4612 2.4612 26.72
      ALPHA_BETA_GAMMA 90. 90. 60.
      SYMMETRY HEXAGONAL
      PERIODIC XYZ
      # and replicate this cell (see text):
      MULTIPLE_UNIT_CELL 6 6 1
    &END CELL
    &TOPOLOGY
      # also replicate the topology (see text):
      MULTIPLE_UNIT_CELL 6 6 1
    &END TOPOLOGY
    &COORD
      SCALED
      # ABC stacked
      C 1./3  1./3  0.
      C 0.    3./3  0.
      C 1./3  1./3  1./8
      C 2./3  2./3  1./8
      C 2./3  2./3  2./8
      C 3./3  0.    2./8
    &END
    &KIND C
      ELEMENT C
      BASIS_SET TZVP-MOLOPT-GTH
      POTENTIAL GTH-PADE-q4
    &END KIND
  &END SUBSYS

&END FORCE_EVAL





--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com<mailto:cp2k+uns... at googlegroups.com>.
To post to this group, send email to cp... at googlegroups.com<mailto:cp... at googlegroups.com>.
Visit this group at https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20180914/b180163b/attachment.html>


More information about the CP2K-user mailing list