[Pw_forum] [QE-GPU] Bug regarding use of INTEL compilers in QE-GPU-5.4.0

[Phanikumar Pentyala]

Dear QE users

I am working on improving properties of doped metal oxides. Installation of
QE-GPU-5.4.0 was done without any errors then I am trying to optimize my
structure (CeO2). After successfully submission of my calculations, either
I am reaching to a stage with *oscillations of total energy *OR calculation
automatically *stops without any error**.* I thought of it's because of BAD
input parameters, so I changed so many parameters as suggested by so many
users in pw_forum like mixing_bata, mixing_dim, smearing, dgauss and based
on exchange correlation functional (NCPP, USPP) I changed ecutwfc and
ecutrho also, but NO use.

After through search, I got a similar previous reported problem (it's very
old): http://qe-forge.org/pipermail/pw_forum/2009-March/086558.html

Same above input file (mine) I tried in 4 processor CPU (desktop) installed
QE-5.4.0. Calculation achieved convergence smoothly.

In the previous thread (link) Prof. Paolo Giannozzi told that it's because
of intel compilers problem and solution for this was linking of local blas.
This I didn't understand, can you explain in elaborated way? These are my
server details

linux: *centOS*

Server: *FUJITSU Server PRIMERGY RX2540 M2*

GPU card:

*NVIDIA Tesla K40 (2 number)*


*intel 2017.1.132 for both mpi and mkl libraries*


*CUDA-8.0.44*
P.S: Input file attached

Thank You



Regards

Phanikumar
Research Scholar
IIT Kharagpur, Kharagpur
West Bengal, India.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20170922/d8d8957d/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ceo2.in
Type: application/octet-stream
Size: 4935 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20170922/d8d8957d/attachment.obj>