[Pw_forum] pw.x memory usage

Pierre-André Cazade pierre.cazade at ul.ie
Mon Jul 7 16:27:45 CEST 2014


Dear all,

I am  new to quantum calculations in periodic systems but I am surprised 
by the amount of memory my QE calculations require.

I am currently working on a study case for transport properties: benzene 
dithiol between gold leads. Due to different surface states for the 
lead, I end up with 268 atoms (256 of gold and 12 for BDT). The 
parameters I am using in my input file are the following:

Au-BDT-Au

  &CONTROL
                  calculation = 'scf' ,
                 restart_mode = 'from_scratch' ,
                       outdir = './outputs' ,
                   pseudo_dir = '/ichec/home/users/cazade/work/lib/pseudo' ,
                       prefix = 'bdt661' ,
                        nstep = 100 ,
  /
  &SYSTEM
                        ibrav = 4,
                    celldm(1) = 32.696678717757,
                    celldm(3) = 2.4036722715606,
                          nat = 268,
                         ntyp = 4,
                      ecutwfc = 22 ,
                      ecutrho = 176 ,
                  occupations = 'smearing' ,
                      degauss = 0.01 ,
                     smearing = 'methfessel-paxton' ,
             exxdiv_treatment = 'gygi-baldereschi' ,
  /
  &ELECTRONS
             electron_maxstep = 100,
                     conv_thr = 1e-8 ,
                  mixing_mode = 'plain' ,
                  mixing_beta = 0.3 ,
  /
ATOMIC_SPECIES
   Au   196.9665   Au.pw91-van_ak.UPF
    C    12.011     C.pw91-van_ak.UPF
    S    32.066     S.pw91-van_ak.UPF
    H     1.008     H.pw91-van_ak.UPF
ATOMIC_POSITIONS angstrom
Au           0.000000000000E+00     0.000000000000E+00 0.000000000000E+00
Au           0.144186143752E+01     0.832459089084E+00 0.235454986781E+01
...
Au 0.230697830003E+02 0.133193454254E+02 36.88
Au 0.245116444378E+02 0.141518045144E+02 39.2346
K_POINTS automatic
   6 6 4   0 0 0

This requires 97GB of RAM when using 2 nodes with 24 cores each (48 
cores in total). Each node has 64GB of memory and yet pw.x is killed 
during the initialization phase without error message from QE. I am well 
aware that the parameters I am using are from being optimal and would 
like to know whether or not it is normal that QE requires such a vast 
amount of memory and, if not, where are the mistakes in my input?

Thank you in advance for you help.

Best regards,
Pierre-Andre Cazade,
University of LImerick



More information about the users mailing list