Author Topic: dscf problem  (Read 5247 times)

kadlubanski

  • Newbie
  • *
  • Posts: 3
  • Karma: +0/-0
dscf problem
« on: May 10, 2013, 09:31:16 am »
Hello,
I have very strange problem with turbomole. It seems that when I run jubex -ri -level mp2 in 8 nodes which 8 cpu each I can get convergence in SCF  but when I start it with 4 nodes 8 cpu each everything goes  correctly. Has anyone similar problem and know the solution ?

Paweł Kadłubański

antti_karttunen

  • Sr. Member
  • ****
  • Posts: 201
  • Karma: +1/-0
Re: dscf problem
« Reply #1 on: May 10, 2013, 07:45:21 pm »
Hi,

how many basis functions does your system have? dscf sometimes has troubles (or at least used to have) if there are too few basis functions per CPU. Since you are using as many as 64 CPUs, you probably should have at least > 1000 AOs?

Antti

uwe

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 420
  • Karma: +0/-0
Re: dscf problem
« Reply #2 on: May 10, 2013, 09:39:04 pm »
Hi,

if you found a reproducible case where dscf fails to converge just because of a certain number of CPUs, you should send that to the support. That's the only way to make sure it will be fixed in a future version.

Regards,

Uwe

antti_karttunen

  • Sr. Member
  • ****
  • Posts: 201
  • Karma: +1/-0
Re: dscf problem
« Reply #3 on: May 11, 2013, 08:44:44 pm »
Oh, true, somehow I missed that Pawel was having convergence issues and not just some crash/hangup.  In this case it must be indeed something else. Anyway, Pawel, here is a very useful FAQ post about some parallelization issues (it's a bit old, but the basic things are certainly still the same): http://www.turbo-forum.com/index.php?topic=78.0;prev_next=prev

Antti