On Nov 23, 2:04 am, mabshoff <[EMAIL PROTECTED]> wrote:
> On Nov 23, 1:57 am, "Alex Ghitza" <[EMAIL PROTECTED]> wrote:
>
> > Hi folks,
>
> > I remember that the question of whether Sage could be ported to CUDA and run
> > on nvidia's GPUs was already brought up on this list.  Here's another
> > incentive to think about it:
>
> >http://tech.slashdot.org/tech/08/11/23/068234.shtml
>
> Well, those numbers are mostly not attainable in real life and it is 4
> GFLops single precision. Double precision is already a whole different
> game and you would be lucky to hit 1 GFlop double precision - not that
> this is anything to sneeze at :)

actually: Peak double precision is 320 GFlops of which you might see
40% in real life. Another big issue is that limited amount of memory
on the GPU itself.

Anyway, sorry for the double post.

> Anyway, Clement and I have gotten FFPACK (which is the core of what
> drives LinBox for example) to run on top of CUDABLAS, but we are
> currently missing Tesla hardware to get the ball rolling in earnest. I
> am also looking into libflame which has the potential to be used as
> GPU scheduler so that all BLAS operations in Sage can be moved utilize
> multiple GPU units. In the end that would accelerate everything that
> uses matrixes of GP(p), charpoly and so on and we are certainly
> working in that direction. Once the new Sage hardware is there we will
> also stuff some GPU hardware in there.
>
> > OK, I'll stop drooling now.
>
> It isn't as powerful as it has been suggested, but for certain
> applications it beats the living crap out of anything else (when
> compared at Flops/Watt or Flops/$)
>
> > Best,
> > Alex

Cheers,

Michael

>
> > --
> > Alex Ghitza -- Lecturer in Mathematics -- The University of Melbourne --
> > Australia --http://www.ms.unimelb.edu.au/~aghitza/
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---

Reply via email to