I expect it will have only a moderate effect. The generated code does currently not use BLAS (although FFC generated BLAS code earlier at some point, currently broken) and matrix insertion does not use BLAS afaik. So assembly should be unaffected. Possibly it might have some effect for the solve step. Will be interesting to hear what you find out.
-- Anders On Fri, Jan 23, 2009 at 02:12:54PM +0200, Evan Lezar wrote: > Hi > > As soon as I have some time, I plan to look into this exact matter. > > Will keep everyone up to date as soon as I get going. > > Evan > > On Fri, Jan 23, 2009 at 1:04 PM, A Navaei <[email protected]> wrote: > > Hi, > > Beside the available optimised CPU-based linear algebra libraries, is > there any profiling available for Dolfin with a GPU-based BLAS, such > as CUBLAS [1]? In this specific case, given the right hardware, > switching to the GPU-based back end is trivial as it comes with the > exact interface of standard BLAS. > > Should we expect a significant boost in Dolfin performance or are > there other remarkable non-BLAS bottlenecks in Dolfin? > > > -Ali > > [1] http://developer.download.nvidia.com/compute/cuda/1_0/ > CUBLAS_Library_1.0.pdf > _______________________________________________ > DOLFIN-dev mailing list > [email protected] > http://www.fenics.org/mailman/listinfo/dolfin-dev > > > _______________________________________________ > DOLFIN-dev mailing list > [email protected] > http://www.fenics.org/mailman/listinfo/dolfin-dev
signature.asc
Description: Digital signature
_______________________________________________ DOLFIN-dev mailing list [email protected] http://www.fenics.org/mailman/listinfo/dolfin-dev
