[QE-users] Quantum-ESPRESSO test run using GPU

[Pietro Bonfa']

Dear Dr. Cantele,

which of the two GPU enabled versions of QE are you testing?
I'm a bit confused by your error message. A quick google search pointed
me to the old qe-gpu-plugin
(https://github.com/fspiga/qe-gpu-plugin/search?utf8=%E2%9C%93&q=something+went+wrong+inside+query_gpu_specs&type=)
developed by Filippo Spiga, but the script in the benchmark repository
that you are considering targets the new porting developed by NVidia and
Spiga (https://github.com/fspiga/qe-gpu).

If you want to try the new version I suggest to use the openmpi bundled
with the PGI compiler and run with 1 MPI process per GPU.

Kind regards,
Pietro Bonfa'


On 03/12/2018 03:50 PM, Giovanni Cantele wrote:
> Dear all,
> 
> I would like to test Quantum-ESPRESSO using GPU. I think I’m not
> properly executing the running command, because I encounter errors like:
> 
> *** ERROR *** something went wrong inside query_gpu_specs! (rank
> 0)-------------------------------------------------------
> Primary job  terminated normally, but 1 process returned
> a non-zero exit code. Per user-direction, the job has been aborted.
> ———————————————————————————
> 
> 
> These are my system specifications:
> i) version: qe-gpu-1.0
> ii) compiler: pgf90 17.10-0 64-bit 
> iii) OpenMPI 3.0.0 compiled with options: ./configure FC=pgfortran
> CC=pgcc CXX=pg++ —with-cuda=/opt/pgi/linux86-64/2017/cuda/8.0
> 
> I’m trying to run the AUSURF112 test, downloaded from the
> benchmark https://github.com/romerojosh/qe-gpu-benchmarks using the
> included run.sh.
> Probably I’m misunderstanding the parameters included therein.
> 
> My small system is composed by two CPUS with 16 cores each, and one
> NVIDIA Quadro P4000 GPU.
> 
> Could you please advise me about how to set the parameters within run.sh?
> 
> #!/bin/bash
> export QE_EXE=/home/cantele/CODES/Quantum-ESPRESSO/qe-5.4.0-GPU/bin/pw-gpu.x
> if [ -z "${QE_EXE}" ]; then echo "ERROR: Please export QE_EXE before
> running script." ; exit; fi
> 
> # QE run parameters
> NGPU=1
> NPOOL=2
> 
> # Node-specific parameters
> GPU_PER_SOCKET=2
> CORES_PER_SOCKET=16
> 
> NCORE_PER_RANK=$((${CORES_PER_SOCKET}/${GPU_PER_SOCKET}))
> 
> export OMP_NUM_THREADS=${NCORE_PER_RANK}
> export MKL_NUM_THREADS=${NCORE_PER_RANK}
> 
> /opt/openmpi/Portland/17.10/bin/mpirun -np ${NGPU} --map-by
> ppr:${GPU_PER_SOCKET}:socket ${QE_EXE} -input ../pw.in -npool ${NPOOL}
> | tee output_${NGPU}GPU_${NPOOL}POOL.txt
> 
> 
> 
> I thank you in advance,
> 
>     Giovanni
> 
> -- 
> 
> Giovanni Cantele, PhD
> CNR-SPIN
> c/o Dipartimento di Fisica
> Universita' di Napoli "Federico II"
> Complesso Universitario M. S. Angelo - Ed. 6
> Via Cintia, I-80126, Napoli, Italy
> e-mail: giovanni.cantele at spin.cnr.it <mailto:giovanni.cantele at spin.cnr.it>
> Phone: +39 081 676910
> Skype contact: giocan74
> 
> ResearcherID: http://www.researcherid.com/rid/A-1951-2009
> Web page: http://people.fisica.unina.it/~cantele
> 
> 
> 
> _______________________________________________
> users mailing list
> users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users
>