[Pw_forum] QE-GPU compilation problems

[Filippo Spiga]

On Sep 28, 2014, at 4:49 PM, Uri Argaman <argamanu at post.bgu.ac.il> wrote:
> 1. You write that there is no benefit in using GPU with small systems (less than 40-50 atoms). What about phonon calculations? 

You are not the first one asking about PHonon and I can tell you will be all pleased at some point in time. I am struggling with time, I have only one pair of hands. Things will happen.

A system made by 40~50 atoms may work with the GPU and give some acceleration. There is no fixed number, I continuously repeat this meme: you have to try (and try on a proper platform).


> 2. We want to do a much more complicated calculation but we do not succeed even with the very simple we write about before.

If you run using not suitable GPU there is a high chance your calculation fails, the reasons can be multiple. I regularly test QE-GPU on GPUs designed for the HPC market, Tesla K20 or K40 (or M20xx series, if you have previous generation cards).


> 3. We try to work with a simple GPU for now, but we intend to work with a better one, but we need to try it first. 

This is a suggestion for all readers, not just for you. DO NOT BUY with your own money any gaming GPU just to try QE-GPU. It is wrong, terrible wrong. 

There are multiple ways you can get a feedback if your calculation suits and get a real benefit from the current QE-GPU implementation:
1. access resources of a national HPC centre: usually national HPC centres have testing or production GPU systems.
2. access NVIDIA GPU test Drive (http://www.nvidia.com/object/gpu-test-drive.html): access is free and QE is one of the codes.
3. send me an input and ask feedback: it looks the easier approach but I do not promise any feedback back in any reasonable time... I am busy, I have little time to handle single requests (with exceptions of real bugs to fix)

It is probably possible to access GPU resources for testing through the HPC Advisory Council network (http://www.hpcadvisorycouncil.com/cluster_center.php). I never tried, I do not know how difficult it is. I can gather information if needed.


> 4. What can we do to run a more complicated calculations on the present GPU?

My answer is embedded in number 3

HTH
F

--
Mr. Filippo SPIGA, M.Sc.
http://filippospiga.info ~ 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."