Author Topic: not enough core memory escf  (Read 6741 times)

e.berardo

  • Jr. Member
  • **
  • Posts: 13
  • Karma: +0/-0
not enough core memory escf
« on: July 10, 2013, 05:24:11 PM »
Dear all,

I am studying the optical properties of a very big system with ca 200 atoms and DZDP basis set.
I want to obtain the 4 lowest excitations as you can see in the input:

Code: [Select]
$scfdamp   start=0.700  step=0.050  min=0.050
$last step      egrad
$scfinstab rpas
$soes all 4
$escfiterlimit 200
$denconv 1d-7
$rpacor 30000

I specified the use of 30 GB of memory, which is almost all the available memory for my machine (31.5 GB).
In the output it is showed the number of vectors that I could study with that amount of memory, which is well above the ones I am studying as you can see below:

Code: [Select]
                      Block Davidson iteration


 total number of roots to be determined:   4


 maximum core memory set to 30000 MB,
 corresponding to      196 vectors in CAO basis


 maximum number of simultaneously treated vectors (including degeneracy):   8


and TM (version 6.4) is able to converge the 4 roots, but then crashes saying that there is not enough memory:

Code: [Select]
   32       a        4        9.241211273980832D-06
 

 converged!


 Switching to fine grid

 preparing numerical integration ....
          Overall gridpoints after grid construction =       1203059

 Calculating ground state density on molecular grid

 Integral of ground state density: N =   1225.998267876637   



 Iteration IRREP Converged      Max. Euclidean
                 roots          residual norm

   33       a        0        1.765535404035553D-04
 

 NOTE: Residual norms are larger than 1.00D-05 due to mgrid option!

 Eigenpairs written on logfile sing_a   



 IRREP   Vector     Eigenvalue           Euclidean residual norm

  a         1    1.403824005910889D-01    1.765535404035553D-04
            2    1.414411435299774D-01    1.705982708125927D-04
            3    1.424511643695836D-01    1.480038908464208D-04
            4    1.429747105057997D-01    1.585635583806746D-04
 

 Not enough core memory!
$rpacor must be set at least to    39

Do you have any idea of why this could be happening,
cheers
Enrico

antti_karttunen

  • Sr. Member
  • ****
  • Posts: 227
  • Karma: +1/-0
Re: not enough core memory escf
« Reply #1 on: July 11, 2013, 11:36:13 AM »
Hi,

are you able to run the job with a smaller amount of memory (e.g. 20000)? 30000 could be a bit too much if your machine has just 32GB. Also, are you running in serial or parallel mode (how many CPUs)? Based on your last error message, one possible scenario that comes to my mind is some kind of integer overflow. You could also try running with the "huge" binaries that should use 64-bit integers everywhere (probably you need to email Turbomole support for these).

Best,
Antti

e.berardo

  • Jr. Member
  • **
  • Posts: 13
  • Karma: +0/-0
Re: not enough core memory escf
« Reply #2 on: July 11, 2013, 11:49:43 AM »
Hi Antti,

good point about the memory usage, I thought about that as well, and I am waiting for the job to start again, but this time with a smaller amount of memory. We'll see!
Anyway, I am running escf with SMP on a node that has 16 cores. I already studied large systems, and never had problems before, this error message confused me quite a lot.
I never considered running the huge turbomole binaries, I'll try to contact the support team!

Thank you for the suggestion,
Enrico
 

e.berardo

  • Jr. Member
  • **
  • Posts: 13
  • Karma: +0/-0
Re: not enough core memory escf
« Reply #3 on: July 12, 2013, 04:44:21 PM »
Update for who could be interested...

The calculation in which I requested a lower amount of memory worked fine. 

Cheers,
Enrico