Questions regarding the compilation of VASP on various platforms: hardware, compilers and libraries, etc.
Moderators: Global Moderator, Moderator
-
call_me_Al
- Newbie
- Posts: 20
- Joined: Tue Oct 04, 2016 7:05 pm
- License Nr.: 5-2273
#1
Post
by call_me_Al » Thu Jul 11, 2019 11:02 am
Hi everyone,
I have vasp.5.4.4 compiled with gfortran on an intel cluster.
The electronic convergence and other features seem to run well.
However, there seems to be a problem when I try to restart from a previously converged WAVECAR, especially is it is too large. I have considered memory etc. but this seems to be ok. (only using about 20% of the node memory). The code usually hangs on a kpoint or at:
Code: Select all
k-point ** : 0.4375-0.4375 0.4375 plane waves: 9258
maximum and minimum number of plane-waves per node : 2352 2283
maximum number of plane-waves: 9306
If I reduce the calculation (no of kpoints for example) it can restart, but is much slower than the original run that produced the WAVECAR. I am not sure if this is a problem with compilation, memory, my physical case or something else. I have compiled successfully with gcc 5.3.0 3.2-gcc5/bin/mpif90
Any ideas welcome.
You do not have the required permissions to view the files attached to this post.
-
merzuk.kaltak
- Administrator
- Posts: 278
- Joined: Mon Sep 24, 2018 9:39 am
#2
Post
by merzuk.kaltak » Mon Mar 23, 2020 3:27 pm
Using gfortran and OpenMPI results in relatively slow VASP binaries. A factor of two (or more) in speed up can be expected using the Intel compiler suite,
respectively the Intel MKL.
Please take a look at following website about supported toolchains for VASP 6,
the GNU+MKL setup works for version 5.4.4 as well:
wiki/index.php/Toolchains