<div dir="ltr"><div dir="ltr">On Mon, Sep 21, 2020 at 8:51 AM Luciano Siqueira <<a href="mailto:luciano.siqueira@usp.br">luciano.siqueira@usp.br</a>> wrote:<br></div><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hi *,<br>
<br>
I'm experimenting with different combinations of KSP solvers and PCs and <br>
I don't know why GMRES/bjacobi are the default choices for CPU and <br>
</blockquote><div><br></div><div>GMRES is chosen because it is monotonic in the 2-norm, and also faster than any other</div><div>Krylov method in terms of iterates if you do not restart. Block-Jacobi is trivially parallel.</div><div>I don't think anyone loves ILU(0), but it is a black-box preconditioner without much overhead,</div><div>so it is the default.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">GMRES/icc are the default choices for GPU. Does anyone know the reason for that?<br></blockquote><div><br></div><div>ICC is only a default if you have a symmetric matrix. I guess you had one on the GPU.</div><div><br></div><div> Thanks,</div><div><br></div><div> Matt</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
Thanks,<br>
Luciano.<br>
</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr" class="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div>What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.<br>-- Norbert Wiener</div><div><br></div><div><a href="http://www.cse.buffalo.edu/~knepley/" target="_blank">https://www.cse.buffalo.edu/~knepley/</a><br></div></div></div></div></div></div></div></div>