[Pw_forum] [Q-e-gpgpu] GPU Amount Memory

[Filippo Spiga]

Dear Rafael,

the appropriate mailing-list for QE-GPU is http://qe-forge.org/mailman/listinfo/q-e-gpgpu I strongly suggest to address any query on that mailing-list directly

On Apr 29, 2013, at 7:29 PM, Rafael Ribadeneira <reribade at unal.edu.co> wrote:
> Dear Forum - Members,
> 
> I request your help with the following error, whereas running pw.x in the gpu-version of QE:
> 
> ---------
> [NEWD] Problem don't fit in GPU memory, memory requested ( 10322157 ) > memory allocated  (0 )!!!
> [VLOC_PSI_K] Problem don't fit in GPU memory --- memory requested ( 75950756 ) > memory allocated  (0 )!!!
> ----------
> 
> We  understand that this is a problem with the amount of memory, we know that we have to modify the file: "cuda_env.h", changing the flag in this file:
> 
> #if defined __CUDA_GET_MEM_HACK
> #define __GPU_MEM_AMOUNT_HACK__
> #endif
> 
> 
> How can we assign the amount of memory by hand in this flag?

Easy answer: you need to specify the amount of memory in byte. 

Complicated answer: your machine or your software environment has something wrong and putting a number in cuda_env.h will probably produce a crash. 

I already exchanged one or two emails with Jarol, he should be aware that I have already suggested that the problem might be your system and not the code. Anyway this "hack" has been deprecated in the 5.0.x. First I strongly suggest you to move to 5.0.x if possible. Based on the information you provide in the email I cannot guess why this happen. That "0" is the amount of free memory on the GPU card detected at runtime by a specific CUDA call. If it return zero there might be several reasons. 

- check that where you compile the CUDA driver in installed (not only the toolkit, also the CUDA driver)
- check that the GPU is visible and works correctly (by running one or two examples provided by the CUDA toolkit)
- check that the GPU is in shared mode (using nvidia-smi)
- check that CUDA is linked properly to the executable and the environment is correct (I can do that if you provide the config.log)
- be sure to use the latest CUDA driver and toolkit PLUS the latest version of QE-GPU

Hope this help

Filippo

--
Mr. Filippo SPIGA, M.Sc.
http://filippospiga.me ~ skype: filippo.spiga

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

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20130502/0d61fcd8/attachment-0001.html>
-------------- next part --------------
_______________________________________________
Q-e-gpgpu mailing list
Q-e-gpgpu at qe-forge.org
http://qe-forge.org/mailman/listinfo/q-e-gpgpu