[mvapich-discuss] Keep some MPI tasks from connecting to GPU

[Devendar Bureddy]

Hi Oliver

We currently have this limitation in supporting heterogeneous mode. We will
fix this issue in future releases.

-Devendar


On Fri, Jun 28, 2013 at 12:49 PM, Oliver Fuhrer (MeteoSwiss) <
oliver.fuhrer at ginko.ch> wrote:

> Hi,
>
> I'm running a hybrid application with 4 MPI tasks restricted to pure CPU
> operation and 8 MPI tasks running compute on GPUs .(amounting to a total of
> 12 MPI tasks per node) I'm running this code on a fat node architecture
> with 8 GPUs connected to two Sandybridge sockets on each node.
>
> Is there a way to keep the pure CPU tasks _completely_ from accessing the
> devices? It seems that upon MPI_Init() all MPI tasks are initializing and
> setting up a context on the device.
>
>
> +-----------------------------------------------------------------------------+
> | Compute processes:                                               GPU
> Memory |
> |  GPU       PID  Process name                                     Usage
>    |
>
> |=============================================================================|
> |    0      7484  ./lm_f90
>  3917MB  |
> |    0      7492  ./lm_f90
>  77MB  |
> |    1      7485  ./lm_f90
>  3922MB  |
> |    1      7493  ./lm_f90
>  77MB  |
> |    2      7486  ./lm_f90
>  3922MB  |
> |    2      7494  ./lm_f90
>  77MB  |
> |    3      7487  ./lm_f90
>  3917MB  |
> |    3      7495  ./lm_f90
>  77MB  |
> |    4      7488  ./lm_f90
>  3909MB  |
> |    5      7489  ./lm_f90
>  3914MB  |
> |    6      7490  ./lm_f90
>  3914MB  |
> |    7      7491  ./lm_f90
>  3909MB  |
>
> +-----------------------------------------------------------------------------+
>
> Cheers,
> Oli
>
> _________
>
> Oliver Fuhrer
> Numerical Models
>
> Federal Departement of Home Affairs FDHA
> Federal Office of Meteorology and Climatology MeteoSwiss
>
> Kraehbuehlstrasse 58, P.O. Box 514, CH-8044 Zurich, Switzerland
>
> Tel. +41 44 256 93 59
> Fax +41 44 256 92 78
> oliver.fuhrer at ginko.ch
> www.meteoswiss.ch - First-hand information
>
>
>
>
>
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>



-- 
Devendar
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20130628/d1c66777/attachment.html