[Pw_forum] huge memory usage in ibrav=1

Lorenzo Paulatto paulatto at sissa.it
Tue Feb 16 12:45:07 CET 2010


On Tue, 16 Feb 2010 12:34:07 +0100, Carlo Nervi <carlo.nervi at unito.it>  
wrote:
> I recently tried to run a simple calculation using a large supercell
> with ibrav=1. The cell contains 13 atoms.
> pw.x runs, but consumes a huge amount of memory (in the example below
> more than 400 Mb of ram for each pw.x process). If I set A = 50.0 it
> consumes more than 1Gb of RAM for each process, and of course having 8
> cPU and 8 Gb of RAM the calculation continously swap with the disk...
> Perhaps I'm doing something wrong, but I cannot find the reason.

Dear Carlo,
the simple-cubic lattice (ibrav=1) is the biggest of them all for constant  
value of alat; furthermore the memory footprint also depends on the  
cutoff, which in your case is very high. If your calculation is well  
converged with only one k-point, you can save half the memory by setting
  K_POINT gamma
which will use arrays of type real, instead of complex, for the  
wavefunctions. Note that this choice is equivalent to
  K_POINTS automatic
   1 1 1   0 0 0
but the latter does not use the Gamma-point specific memory-saving tricks.  
On the other hand it is *not* equivalent to
  K_POINTS automatic
   1 1 1   1 1 1
which is normally more converged.

Finally, I haven't specified the parallelization paradigm you are using;  
please not that memory is *not* distributed among pools. If your situation  
is really critical you may wish to use only 4 or even 2 process together  
with openmp (thread-based parallelization) which should ensure a better  
memory scalability. I think this option is only available in the CVS  
version of QE.

best regards

-- 
Lorenzo Paulatto

*** Note: my affiliation has changed! please send future
correspondence to: <Lorenzo.Paulatto at impmc.upmc.fr> ***

post-doc @ IMPMC/UPMC - Université Paris 6
phone: +33 (0)1 44 27 74 89
www:   http://www-int.impmc.upmc.fr/~paulatto/

previously:
phd student @ SISSA  &  DEMOCRITOS (Trieste)
phone: +39 040 3787 511
www:   http://people.sissa.it/~paulatto/



More information about the users mailing list