Knowing how much memory exactly a code uses is close to impossible. There are some explanations about memory usage in the documentation: http://www.quantum-espresso.org/wp-content/uploads/Doc/pw_user_guide/node13.html The code prints out a list of the presumably largest arrays (it also prints out, for some machines and compilers, the amount of dynamically allocated arrays, not sure how reliable). This gives you at least a lower limit on how much memory is needed (per process: on a multi-core machine, multiply by the number of cores to have the total RAM). Note that using "Gamma tricks" reduces a lot the amount of needed memory
Paolo On Thu, Apr 14, 2016 at 10:03 AM, Elad Segev <[email protected]> wrote: > Dear QE users and developers, > > I running a pw 'relax' calculation in a unit cell with 512 atoms of GeSe > with fcc symmetry and am running on a machine with 64 cores and 512GB RAM. > I encountered difficulties when tried to run on all cores (64) and the > memory usage was at 100% and started using the swap file heaviliy whille > most of the cores are not running with full CPU but only ~1% . > > After trying to run on less cores on the macine, I ran on 16 cores and > found that the memory usage was at 502GB and with no use of the swap file, > all the cores are running at 100% CPU. > > I'm runinng this job with 1 k-point so npool=1. > the MPI command is: > mpirun pw.x -np 16 -npool 1 -nband 1 -ntg 1 > opmthreads=1 > > Is there a solution in which I can run on all cores? > > Thanks in advance, > > Elad Segev, > Ph.D. candidate > Ilse KAtz Institute of Nanoscale science and technology > Ben-Gurion University of the Negev, Israel > > _______________________________________________ > Pw_forum mailing list > [email protected] > http://pwscf.org/mailman/listinfo/pw_forum > -- Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche, Univ. Udine, via delle Scienze 208, 33100 Udine, Italy Phone +39-0432-558216, fax +39-0432-558222
_______________________________________________ Pw_forum mailing list [email protected] http://pwscf.org/mailman/listinfo/pw_forum
