Author Topic: em64t and $parallel_platform  (Read 11334 times)

antti_karttunen

  • Sr. Member
  • ****
  • Posts: 227
  • Karma: +1/-0
em64t and $parallel_platform
« on: April 10, 2007, 11:35:14 AM »
I mentioned this issue in a post made in ricc2 subforum, but maybe its better to report it here, too.

I noticed that in Turbomole 5.9 the $parallel_platform keyword is always set to "cluster" when running parallel calculations with em64t architechture. Would "$parallel_platform MPP" be more appropriate, considering that mpirun_scripts use MPP for all the other x86 and x86-64 based architechtures?

christof.haettig

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 291
  • Karma: +0/-0
    • Hattig's Group at the RUB
Re: em64t and $parallel_platform
« Reply #1 on: March 15, 2008, 12:43:49 PM »
these keywords have (to the best of my knowledge) only an influence on the diagonalization in SCF and DFT calculations. They have no effect on ricc2 runs.

Christof

uwe

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 558
  • Karma: +0/-0
Re: em64t and $parallel_platform
« Reply #2 on: March 17, 2008, 09:41:07 AM »
Hi,

there was a line missing in the scripts that prepare the parallel input. $parallel_platform only controls the number of CPUs used in parallel linear algebra steps, and switches on ScaLAPACK when choosing SMP. Patch files and a new package for Linux/PC will be available soon.

Uwe