[Pw_forum] A "relax" input runs on CPU (pw.x) but not on CPU-GPU (pw-gpu.x)

[Filippo Spiga]

Dear Reza Behjatmanesh-Ardakani,

On Jun 14, 2014, at 8:38 PM, Reza Behjatmanesh-Ardakani <reza_b_m_a at yahoo.com> wrote:
> I have sent this error to the QE-GPU forum nearly 2 weeks ago; however, I didn't receive any reply. Maybe in QE forum one can help, so I repeat here:

I haven't noticed the message, it is possible that my email client put it into SPAM automatically. Apologize.


> I parepared an input for relaxation of carbon nanotube that
> is in periodic form only in one-dimension with carbon
> monoxide in it.
> While it runs with pw.x, it encounter with error by pw-gpu.
> Both programs are on the same machine, but in different
> directory ( pw.x is in espresso-5.0.3, pw-gpu is in
> espresso-503).
> I installed the patch files on both programs. I used MPICH
> and ifort (comoserxe 2013 update 2) for both.
> I compiled pw-gpu.x without magma, as I saw some errors
> such as cdiaghg error related to it.
> I used core-i7 and GTX-660Ti, all for test. I used pw-gpu.x
> on this system to run 1*1*3 supercell of carbon nanotube
> without any problem.
> The only difference of this new input with its pure form
> comes back to the presence of CO inside it and vacuum in x
> and y direction.
> My GPU has only 2G and my system has 8G (the system is only
> for test). 

If you run successfully one input case but another test-case does not work maybe the new physical system you want to simulate is too big (some data structure cannot fit GPU memory) or it has some unsupported features.

Anyway, I do _not_ suggest to use a gaming card for QE-GPU. I understand how appealing is to use gaming cards to perform computation but QE-GPU is developed, maintained and tested for NVIDIA TESLA cards (Fermi or Kepler generation). It works, in principle, using gaming cards (GTX). However I do not guarantee performance or results reproducibility. At some point in time I will insert a control in the code such has if someone is trying to run the code on un-supported GPU, the code will terminate before start any calculation.  

Regards,
Filippo

--
Mr. Filippo SPIGA, M.Sc.
http://www.linkedin.com/in/filippospiga ~ skype: filippo.spiga

«Nobody will drive us out of Cantor's paradise.» ~ David Hilbert

*****
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."