[Pw_forum] Numeric of

Filippo Spiga spiga.filippo at gmail.com
Tue Aug 4 09:11:17 CEST 2015


Dear Cheung,

do not use the public version of QE-GPU 14.x for spin magnetization calculation. I am aware of this problem that appear in some corner cases since a while but the code does not automatically warn people they are running using an unsupported feature. Maybe in the future unsupported feature need to trigger an explicit ABORT to warn users.

Use the CPU version for now, the problem will be fixed soon.

F

On Aug 3, 2015, at 6:58 PM, Cheung, Samson H. (ARC-TN)[Computer Sciences Corporation] <samson.h.cheung at nasa.gov> wrote:
> Sorry, I have not finished the email!!
> Let me continue:
> 
> 
> pw_cpu.out:
>      Magnetic moment per site:
>      atom:    1    charge:    1.6134    magn:   -0.0000    constr:    0.0000
>      atom:    2    charge:    1.6136    magn:    0.0000    constr:    0.0000
>      atom:    3    charge:    1.6135    magn:   -0.0000    constr:    0.0000
>      atom:    4    charge:    1.6136    magn:   -0.0000    constr:    0.0000
>> 
> pw_gpu.out:
>      Magnetic moment per site:
>      atom:    1    charge:    2.0472    magn:   -0.7654    constr:    0.0000
>      atom:    2    charge:    2.0434    magn:   -0.7580    constr:    0.0000
>      atom:    3    charge:    2.0454    magn:   -0.7623    constr:    0.0000
>      atom:    4    charge:    2.0426    magn:   -0.7574    constr:    0.0000
>> 
> Do you have any theory about the numerical difference between the CPU and GPU case?
> Many thanks!!
> 
> ~Samson
> 
> 
> From: "Cheung, Samson H. (ARC-TN)[Computer Sciences Corporation]" <samson.h.cheung at nasa.gov>
> Date: Monday, August 3, 2015 at 10:56 AM
> To: "spiga.filippo at gmail.com" <spiga.filippo at gmail.com>
> Cc: "pw_forum at pwscf.org" <pw_forum at pwscf.org>
> Subject: Numeric of 
> 
> 
> 
> 
> I downloaded GRIR443 case from http://qe-forge.org/gf/project/q-e/frs/?action=FrsReleaseBrowse&frs_package_id=36
> 
> The way I compile the CPU case is follow:
> %> setenv FC ifort
> %> setenv LDFLAGS -lmpi
> %> setenv CC icc
> %> setenv FCFLAGS "-O3 -axCORE-AVX2 -xSSE4.2 -assume byterecl"
> %>  ./configure --prefix=/nobackup/scheung/espresso LIBDIRS="-mkl"
> 
> The way I get and compile the GPU add-on is follow:
> %> wget https://github.com/fspiga/QE-GPU/archive/v14.10.0.tar.gz
> %> tar zxvf v14.10.0.tar.gz
> %> mv QE-GPU-14.10.0/GPU espresso-5.1.2/
> %> cd espresso-5.1.2/GPU/
> %> module load cuda/6.5
> %> cd GPU/
> %> ./configure --enable-parallel --enable-openmp --enable-cuda --with-gpu-arch=sm_35 --with-cuda-dir=/nasa/cuda/6.5 --without-magma --with-phigemm  --without-scalapack
> %> cd ..
> %> make -f Makefile.gpu -j4 pw-gpu
> 
> 
> The  way I ran the codes is follow:
> mpiexec -np 80 ./pw-gpu.x  -input grir443.in > pw_gpu.out
> mpiexec -np 80 ./pw.x  -input grir443.in > pw_cpu.out
> 
> However, I see the differences between the numeric reported from pw_gpu.out and pw_cpu.out:
> 
> pw_cpu.out:
>      Magnetic moment per site:
>      atom:    1    charge:    1.6134    magn:   -0.0000    constr:    0.0000
>      atom:    2    charge:    1.6136    magn:    0.0000    constr:    0.0000
>      atom:    3    charge:    1.6135    magn:   -0.0000    constr:    0.0000
>      atom:    4    charge:    1.6136    magn:   -0.0000    constr:    0.0000
> 
> 
> 

--
Mr. Filippo SPIGA, M.Sc.
Quantum ESPRESSO Foundation
http://fspiga.github.io ~ skype: filippo.spiga

*****
Disclaimer: "Please note this message and any attachments are CONFIDENTIAL and may be privileged or otherwise protected from disclosure. The contents are not to be disclosed to anyone other than the addressee. Unauthorized recipients are requested to preserve this confidentiality and to advise the sender immediately of any error in transmission."





More information about the users mailing list