cp2k on CRAY-XT5 rosa

[vweber]

Hi Carlo

Running your input (only the first 6 SCF step) on blanc (16 prcs) gave
me the following timings

655.19 s (cp2k from 17.8.09)
692.53 s (cp2k from 6.11.09)

there is clearly a slow down in the dbcsr_sm_fm_multiply that we are
working on.
I suggest you to use the other OT which is faster, i.e.

      &OT
        PRECONDITIONER  FULL_SINGLE_INVERSE
        ALGORITHM IRAC
        ORTHO_IRAC CHOL
        EPS_IRAC_SWITCH 0.0E0
        MINIMIZER  CG
      &END

with that I get

602.71 s (cp2k from 6.11.09)

... even with about 40 s slowdown in the dbcsr_sm_fm_multiply routine.

V




On Nov 6, 10:31 am, carlo antonio pignedoli <c.pig... at gmail.com>
wrote:
> Dear all,
> I use cp2k on the CRAY-XT5 of CSCS in Manno.
>
> Yesterday I recompiled the program and i noticed a considerable slowing down
> for GPW.
>
> I was running
>   **** **** ******  **  PROGRAM STARTED AT               Thu Nov  5 11:26:04
> 200
>  CP2K| version string:                  CP2K version 2.0.1 (Development
> Version)
>  CP2K| is freely available fromhttp://cp2k.berlios.de/
>  CP2K| Program compiled at                         Wed Jul  1 22:46:48 CEST
> 2009
>  CP2K| Program compiled on
> rosa3
>  CP2K| Program compiled for
> CRAY-XT5
>
> Compiled with the CRAY-XT5.popt included in the sistribution and with
> module swap  PrgEnv-pgi/2.2.41  PrgEnv-pathscale/2.2.21A
> module load fftw/3.2.2
>
> ~400 atoms on 128 cores     ABC [angstrom]  23.50847160 22.621037531168 30
> &FORCE_EVAL
>   METHOD Quickstep
>   &DFT
>     BASIS_SET_FILE_NAME ../BASIS_MOLOPT
>     POTENTIAL_FILE_NAME ../GTH_POTENTIALS
>     RESTART_FILE_NAME ./DH-RESTART.wfn
>     &QS
>       METHOD GPW
>       EXTRAPOLATION ASPC
>       EXTRAPOLATION_ORDER 3
>     &END QS
>     &MGRID
>       CUTOFF 280
>       NGRIDS 5
>     &END
>     &SCF
>       MAX_SCF 30
>       SCF_GUESS RESTART
>       EPS_SCF 2.0E-6
>       &OT
>         PRECONDITIONER  FULL_SINGLE_INVERSE
>         MINIMIZER  CG
>       &END
>       &OUTER_SCF
>         MAX_SCF 15
>         EPS_SCF 2.0E-6
>       &END
>
>      1 OT CG       0.15E+00   28.3     0.00001269    -17545.3799531547
> -1.75E+04
>      2 OT LS       0.11E+00    4.6     1.00000000    -17545.3801164098
> -1.63E-04
>      3 OT CG       0.11E+00   11.4     0.00001158    -17545.3801385692
> -2.22E-05
>
> I updated and compiled again with CRAY-XT5.popt and now I get, for the same
> run
>
>   **** **** ******  **  PROGRAM STARTED AT               2009-11-05
> 17:43:03.925
>  CP2K| version string:                 CP2K version 2.1.61 (Development
> Version)
>  CP2K| is freely available fromhttp://cp2k.berlios.de/
>  CP2K| Program compiled at                          Thu Nov  5 15:54:57 CET
> 2009
>  CP2K| Program compiled on
> rosa3
>  CP2K| Program compiled for
> CRAY-XT5
>
>      1 OT CG       0.15E+00   41.7     0.00001164    -17545.3823332203
> -1.75E+04
>      2 OT LS       0.12E+00    6.7                   -17545.3824862257
>      3 OT CG       0.12E+00   15.4     0.00000988    -17545.3824983556
> -1.65E-04
>
> Was there any major change in the code or should I check with the system
> management if
> something changed in the libraries installed on the machine?
>
>  Thanks a lot for your help
>
> Carlo