[Pw_forum] better phonon parallelization
adwait mevada
adwait.mevada at gmail.com
Thu Sep 3 21:56:20 CEST 2015
Dear QE users,
I am running phonon calculations on Intel Xeon W3520 processor
it has 8 cores @2.66GHz with 16gb ram
I am currently doing phonon runs for mgzn2 system
initially my command for running ph.x for 4 processors
mpirun -np 4 ph.x -ndiag 2 -in mgzn2.in > mgzn2.out
which had following details in the out file
Parallel version (MPI), running on 4 processors
R & G space division: proc/nbgrp/npool/nimage = 4
...
Alpha used in Ewald sum = 2.8000
PHONON : 9h12m CPU 9h18m WALL
observing that band structure calculation it self has taken 9hr
the whole calculation would take much longer time.
so i decided to use all 8 cores
mpirun -np 8 ph.x -nk 2 -in mgzn2-ph.in > mgzn2-ph.out
Parallel version (MPI), running on 8 processors
R & G space division: proc/nbgrp/npool/nimage = 8
...
Alpha used in Ewald sum = 2.8000
PHONON : 5h18m CPU 5h21m WALL
does any one have an idea as to what possible combination
would ensure a much faster calculation ?
I also have a question regarding k points calcuated in phonons.
while doing scf calculations for mgzn2 the number of k points
were 60, but for the same input to phonon calculations the
number of k points increased to 312 why is that the case?
lastly for the phonon calculation the number of symmetries found is 24
along with inversion but for q point calculation the symmetries reduce to 9?
Isnt the brillouin zone for hcp structure also an hcp?
24 Sym. Ops., with inversion, found
Computing dynamical matrix for
q = ( 0.0000000 -0.5773503 0.0000000 )
9 Sym.Ops. (with q -> -q+G )
Thanks & Regards,
Adwait Mevada,
Ph.D. Student,
Gujarat University,
India.
--
-Adwait
'I Wonder!...The Empty Cup'
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20150904/29001e02/attachment.html>
More information about the users
mailing list