Update on PETSc GPU

[Matthew Knepley]

On Thu, Aug 20, 2009 at 8:48 AM, Stephen Ball <Stephen.R.Ball at awe.co.uk>wrote:

> Hi
>
> We are currently looking at procuring a GPU cluster for testing in the
> near future. I would be very keen to test your PETSc GPU when it is
> available. Can you give me an update on when you plan to make PETSc GPU
> available?


I don't think before the end of the year, because there is a lot to get
done. I will
post a repo soon, so you can always get the unfinished business just like
petsc-dev.
There is not a lot of PETSc that works well on a GPU, however, so I just
want to
list the main concerns:

  a) Sparse MV: Not great, and we plan to use the NVIDIA implementation

  b) Triangular Solve: Ditto, although we might rewrite some of it

  c) FMM: I would like to put in an interface to PetFMM-GPU

  d) Finite element integrals/stencils: There benefit greatly from GPU
      acceleration, but is not part of PETSc. Not sure how to do this, but
      we would work with users.

  e) Factorization: I hate this, but so many people do it. You probably only
need
      the triangular solve part here, so it comes down to b)

  f) MG: This is jsut a bunch of a) and b) to organize

d) is the most problematic, since its very important, but would demand a lot
of
user work I anticipate.

  Matt


> Regards
>
> Stephen R. Ball
>
> This e-mail and any attachments may contain confidential and
> privileged information. If you are not the intended recipient,
> please notify the sender immediately by return e-mail, delete this
> e-mail and destroy any copies. Any dissemination or use of this
> information by a person other than the intended recipient is
> unauthorized and may be illegal.
>
-- 
What most experimenters take for granted before they begin their experiments
is infinitely more interesting than any results to which their experiments
lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20090826/d539705e/attachment.htm>