<div dir="ltr">Okay, thank you for the guidance. <div>Jim</div></div><div class="gmail_extra"><br><br><div class="gmail_quote">On Mon, Aug 12, 2013 at 1:38 PM, Karl Rupp <span dir="ltr"><<a href="mailto:rupp@mcs.anl.gov" target="_blank">rupp@mcs.anl.gov</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi Jim,<br>
<br>
in addition to what Matt already said, keep in mind is that you usually won't see a two-fold performance gain in iterative solvers anyway, as the various integers used for storing the nonzeros in the sparse matrix don't change their size. I once played with an implementation of an non-preconditioned mixed-precision CG solver, and I only obtained about a 40 percent overall performance gain for well-conditioned systems. For less well-conditioned systems you may not get any better overall performance at all (or worse, fail to converge).<br>
<br>
Best regards,<br>
Karli<div class="im"><br>
<br>
<br>
On 08/12/2013 12:32 PM, Matthew Knepley wrote:<br>
</div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="im">
On Mon, Aug 12, 2013 at 12:24 PM, Jim Fonseca <<a href="mailto:jefonseca@gmail.com" target="_blank">jefonseca@gmail.com</a><br></div><div class="im">
<mailto:<a href="mailto:jefonseca@gmail.com" target="_blank">jefonseca@gmail.com</a>>> wrote:<br>
<br>
Hi,<br>
We are curious about the mixed-precision capabilities in NEMO5. I<br>
see that there is a newish configure option to allow single<br>
precision for linear solve. Other than that, I found this old post:<br>
<a href="https://lists.mcs.anl.gov/mailman/htdig/petsc-users/2012-August/014842.html" target="_blank">https://lists.mcs.anl.gov/<u></u>mailman/htdig/petsc-users/<u></u>2012-August/014842.html</a><br>
<br>
Is there any other information about to see if we can take advantage<br>
of this capability?<br>
<br>
<br>
Mixed-precision is hard, and especially hard in PETSc because the C type<br>
system is limited.<br>
However, it also needs to be embedded in an algorithm that can take<br>
advantage of it. I would<br>
always start out with a clear motivation:<br>
<br>
- What would mixed precision accomplish in your code?<br>
<br>
- What is the most possible benefit you would see?<br>
<br>
and decide if that is worth a large time investment.<br>
<br>
Thanks,<br>
Jim<br>
<br>
--<br>
Jim Fonseca, PhD<br>
Research Scientist<br>
Network for Computational Nanotechnology<br>
Purdue University<br></div>
<a href="tel:765-496-6495" value="+17654966495" target="_blank">765-496-6495</a> <tel:<a href="tel:765-496-6495" value="+17654966495" target="_blank">765-496-6495</a>><br>
<a href="http://www.jimfonseca.com" target="_blank">www.jimfonseca.com</a> <<a href="http://www.jimfonseca.com" target="_blank">http://www.jimfonseca.com</a>><div class="im"><br>
<br>
<br>
<br>
<br>
<br>
--<br>
What most experimenters take for granted before they begin their<br>
experiments is infinitely more interesting than any results to which<br>
their experiments lead.<br>
-- Norbert Wiener<br>
</div></blockquote>
<br>
</blockquote></div><br><br clear="all"><div><br></div>-- <br>Jim Fonseca, PhD<div>Research Scientist</div><div>Network for Computational Nanotechnology</div><div>Purdue University</div><div>765-496-6495<br><div><a href="http://www.jimfonseca.com" target="_blank">www.jimfonseca.com</a></div>
<div><br><div><br></div></div></div>
</div>