[Pw_forum] QE-GPU with Nvidia K80 failed

[Rolly Ng]

Hello Filippo,

Thank you!

That was what I did, I configure it with

./configure CC=icc F90=ifort F77=ifort MPIF90=mpiifort \
--enable-parallel --disable-openmp --without-scalapack \
--enable-cuda --with-gpu-arch=sm_35 --with-cuda-dir=/usr/local/cuda-6.5 \
--with-magma --with-phigemm \
LDFLAGS="-L/usr/lib64/ -lstdc++"

But, as I run,
mpirun -genvall -np 8 ./pw-gpu.x -inp 
/home/rolly/AuSurf111_fix/AuSurf111_fix+Biotin.in |tee 
/home/rolly/AuSurf111_fix/AuSurf111_fix+Biotin.out

It stops immediately.

*** ERROR *** something went wrong inside query_gpu_specs! (rank 0)*** 
ERROR *** something went wrong inside query_gpu_specs! (rank 2)*** ERROR 
*** something went wrong inside query_gpu_specs! (rank 1)*** ERROR *** 
something went wrong inside query_gpu_specs! (rank 3)*** ERROR *** 
something went wrong inside query_gpu_specs! (rank 4)*** ERROR *** 
something went wrong inside query_gpu_specs! (rank 7)*** ERROR *** 
something went wrong inside query_gpu_specs! (rank 5)*** ERROR *** 
something went wrong inside query_gpu_specs! (rank 
6)linux-iltf:/opt/QE520-GP
U/bin #

I have to do control-c to quit.

Any idea? Thank you!

Rolly

On 01/09/2016 06:00 PM, Filippo Spiga wrote:
> Hello Rolly,
>
> Please try to use "--with-gpu-arch=sm_35", compute capability 3.5 is supported for K80. I do not recall I added explicit support for 3.7 in QE-GPU. From an optimization perspective, I can guarantee there is nothing in the code that leverage K80 better than any other NVIDIA GPU Kepler architecture.
>
> Best suggestion is to move to QE 5.3.0 and get the QE-GPU bundled with it. Updates will arrive with the next release as well since now the two are aligned. tar.gz and announcement will made during the weekend.
>
> Cheers
>
> --
> Mr. Filippo SPIGA, M.Sc.
> Quantum ESPRESSO Foundation
> http://www.quantum-espresso.org ~ skype: filippo.spiga
>
>
>> On Jan 9, 2016, at 9:28 AM, Rolly Ng <rollyng at gmail.com> wrote:
>>
>> Dear QE-GPU developers,
>>
>> I am testing the latest NV K80 for QE-GPU, and the test bed has 4x K80,
>> total of 8 GPU cores.
>>
>> I was successful using QE-GPU on C2050 with sm_20, but the K80 comes
>> with sm_37.
>>
>> I have cuda-6.5 installed and I am using QEv5.2.0 + QE-GPUv14.10.0
>> complied okay. However, when I run the pw-gpu.x, it failed with
>> *** ERROR *** something went wrong inside query_gpu_specs! (rank 0) ...
>> (rank 7)
>>
>> Should I upgrade to cuda 7.0 or 7.5 and QEv5.3.0 + QE-GPUv5.3?
>>
>> Thank you,
>> Rolly
>>
>> -- 
>> PhD. Research Fellow,
>> Dept. of Physics & Materials Science,
>> City University of Hong Kong
>> Tel: +852 3442 4000
>> Fax: +852 3442 0538
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://pwscf.org/mailman/listinfo/pw_forum
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum

-- 
PhD. Research Fellow,
Dept. of Physics & Materials Science,
City University of Hong Kong
Tel: +852 3442 4000
Fax: +852 3442 0538