Increasing nb of cores per node degrade drastically the performance of cp2k.popt

[Shadrina Maria]

Dear All,

I've started to use cp2k and noticed that job is slowed down on 2x8 procs in
comparison with 2x2.
We compared CPU timing per opt step for QM/MM task:

 nodes x cores/node   CPU time per iter, sec

2x2          115

1x4          108

4x2            70

1x8            82

2x8         4500

4x8         4200

Using of TRACE setting revealed that problem is mainly in cp_fm_syevd_base
subroutine which spent almost all of this additional CPU time in 2x8 and 4x8
cases.




And we also compared CPU times of test 32H2O-md.inp for 2x2 and 2x8 cases
and got:

 nodes x cores/node   CPU time per iter, sec

2x2          33

2x8          172

TRACE revealed that qs_forces,  qs_energies_scf,  scf_env_do_scf,
velocity_verlet,  qs_forces,  qs_energies_scf,   scf_env_do_scf and others
are 5 times slower in 2x8 than 2x2


Do you have any suggections and ideas why it's happened?


CP2K version 2.2.262
the lib is MKL-Scalapack
the system is cluster of XEON nodes (8cores/node) with Infiniband switch
the compiler is Intel's

Thank you in advance.
Best regards, Maria.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20110712/0f90e79a/attachment.htm>