Re: [Starpu-devel] CUDA Allocate problem

[Berenger Bramas <berenger.bramas@inria.fr>]

Hi, thanks for the support,

By executing:
STARPU_LIMIT_CUDA_MEM=400 STARPU_NCUDA=1 ./Tests

I still have the error : CUDA Error: out of memory
But the GPU should have more than 6GB available.

Thanks.

Bérenger Bramas

HiePACS Project

Tel (05 24 57) 40 76
INRIA BORDEAUX Sud Ouest


----- Mail original -----
| De: "Samuel Thibault" <samuel.thibault@ens-lyon.org>
| À: "Berenger Bramas" <berenger.bramas@inria.fr>
| Cc: starpu-devel@lists.gforge.inria.fr, "Emmanuel Agullo" 
<emmanuel.agullo@inria.fr>
| Envoyé: Mardi 26 Mars 2013 12:55:30
| Objet: Re: CUDA Allocate problem
| 
| Berenger Bramas, le Tue 26 Mar 2013 12:50:12 +0100, a écrit :
| > "oops in allocate_vector_buffer_on_node (datawizard/interfaces/
| > vector_interface.c:264)... 4: unspecified launch failure"
| 
| Interesting error from CUDA...
| 
| It looks like we can not really rely in CUDA to be able to tell
| us when it does not have memory any more... You can try to set
| STARPU_LIMIT_CUDA_MEM to the amount of MiB you want to let StarPU use
| before making room without waiting for CUDA to report errors.
| 
| Samuel
|