Dear all,

I am performing PwScf for 5x5x5 Vanadium supercell (containing 125 atoms)
with ifort compiled pw.x on a cluster with 1 node of max 64 cores of AMD
CPUs  (128 GB RAM).
Surprisingly, my primary results show that the calculations with 24 cores
are much faster than 64 ones (total time to reach 5 iterations); 12 cores
faster than 48 cores and so on (there is no semi-linear increase in
parallel performance). Also, it takes a really long time to reach
convergence within a very light criterion (e.g., 1e-3 electron optimization
and 49 Ry cutewf). Moreover, when I import defects such as vacancies or
interstitials, the calculations do not converge at all (with 100
iterations).
I am completely out of any idea how to fix these issues and do not know
whether I have failed in my primary compilation or there are some options I
miss that can boost the performance and reduce the calculations time.
Can anyone give me a recommendation?
Also I have attached my input file.


Bests,
Afshin
-- 
With Best Regards
Afshin Arjhangemehr
PhD candidate in Radiation Application

Attachment: V_125atom_primi_trigonal.in
Description: Binary data

_______________________________________________
Pw_forum mailing list
[email protected]
http://pwscf.org/mailman/listinfo/pw_forum

Reply via email to