[Trilinos-Users] Trilinos on GPU?

[Heroux, Michael A]

Gianluca,

Trilinos supports GPUs via the Kokkos library.  Support is limited right now, but growing quickly, including multinode support.  Trilinos 10.4, coming out in a few weeks, will have a lot more.  The links Ted mentioned will work well for you on a single GPU.

Right now, for general sparse computations, multinode GPU computations are not competitive with multinode CPU performance because of host-device data transfer costs, although there are certainly specialized cases where this is not true.  The situation will improve over the next year, at which time we will have support for scalable multinode GPU, and hybrid GPU/CPU capabilities ready in Trilinos.

Mike


On 6/25/10 12:55 PM, "Ted Kord" <teddy.kord at gmail.com> wrote:

Have a look at CUSP (http://code.google.com/p/cusp-library/), ViennaCL and Magma.

Ted


On 25 June 2010 18:36, Gianluca Moro <giangiammy at gmail.com> wrote:
Hi all,

I'm working on GPU and as PhD project I'm looking around
for automatic optimization of Linear Algebra routines running
on GPU (something like ATLAS).

Is there any project in this sense, or there's any idea to
add such a support in Trilinos packages?

thanks
Gianluca
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://software.sandia.gov/pipermail/trilinos-users/attachments/20100625/e111bc95/attachment.html