[petsc-dev] CUDA sm_13 for double

[Filippo Spiga]

Dear Barry,
    I guess the best is to probe directly the GPUs instaled on the system
using the NVIDIA CLI tool called nvidia-smi. Command line options "-a" or
"-q" report the model. The problem will be parse the information in the
sdtout. Maybe we can ping NVIDIA to improve the output to be more
"machine-readable". It changes if drivers change. It can be very annoying (I
am speaking for direct experience)....

Regards

--
Filippo SPIGA, MSc Computer Science
~ homepage: http://tinyurl.com/fspiga ~

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


On Tue, Jan 11, 2011 at 11:44 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:

>
> On Jan 11, 2011, at 4:55 PM, Lisandro Dalcin wrote:
>
> > I think the two lines below (config/PETSc/packages/cuda.py ) are wrong:
> >
> >      if self.scalartypes.precision == 'double':
> >        self.setCompilers.addCompilerFlag('-arch sm_13')
> >
> > What if your GPU is sm_20?
>
>    This is a hack to get things to work. We've love for you to tell us the
> correct solution. Should it try to set the arch to the highest one supported
> by the system? (If so how do we find out the highest)?
>
>  Thanks
>
>   Barry
>
>
> >
> > --
> > Lisandro Dalcin
> > ---------------
> > CIMEC (INTEC/CONICET-UNL)
> > Predio CONICET-Santa Fe
> > Colectora RN 168 Km 472, Paraje El Pozo
> > Tel: +54-342-4511594 (ext 1011)
> > Tel/Fax: +54-342-4511169
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110112/60fb9d9e/attachment.html>