[Trilinos-Users] Fwd: Epetra vs. Tpetra: tridiagonal matrix speedup

[Travis Austin]

Hi Nico,

What were the absolute timing differences at low processor count between Epetra and Tpetra?
I'm working on some timing diffs between Epetra and the new version that we are working on
that will yield 64 bit int capability?  I'm trying to get a handle on the cost that will be imposed on
our approach so I am comparing to the original Epetra implementation and Tpetra.

Any info you can provide on absolute timing diffs between Epetra and Tpetra would be of great
help to me.

Thanks,
Travis


^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Travis Austin, Ph.D.
VP of Computational Mathematics
Tech-X Corporation
5621 Arapahoe Ave, Suite A
Boulder, CO 80303
austin at txcorp.com
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^



On Apr 16, 2012, at 12:44 PM, Nico Schlömer wrote:

> Hi all,
> 
> For anyone interested:
> I was curious how Epetra compares to Tpetra in the latest dev::master,
> and ran some simple scaling tests this morning with the matrix
> diag(-1,2,-1) for matrix construction and apply(). The dimension of
> the system is 10M and the maximum number of procs tested a tiny 128
> (oh well).
> 
> It seems I cannot attach pictures here, so I just put the results at
> <https://win.ua.ac.be/trilinos/content/epetra-vs-tpetra-tridiagonal-matrices>.
> 
> It turns out that Epetra still has the upper hand considering absolute
> timings (not shown) as well as strong scaling (see figures).
> For both shared and distributed memory machines, the speedup
> is around a not too impressive 0.5 for all numprocs tested, but at
> least it doesn't decrease too much with increasing numprocs.
> 
> Cheers,
> Nico
> 
> _______________________________________________
> Trilinos-Users mailing list
> Trilinos-Users at software.sandia.gov
> http://software.sandia.gov/mailman/listinfo/trilinos-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://software.sandia.gov/pipermail/trilinos-users/attachments/20120501/43734493/attachment.html