<div dir="ltr"><div class="gmail_extra"><br><div class="gmail_quote">On Tue, Feb 12, 2013 at 7:09 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"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="im">Which crossover are you referring to? The CPU versus GTX285 at about 20k<br>
dofs, but with only very small gains for another order of magnitude?<br>
</div></blockquote>
<br>
I assume it's the cross-over of Xeon Phi vs. Xeon,</blockquote><div><br></div><div style>MIC is slower than Xeon by more than an order of magnitude at 10k dofs.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
but almost all cross-overs happen in the 10k-100k region and are due to PCI-Express latency. </blockquote><div><br></div><div style>Why is PCI-Express latency important here? Can't the MIC code run entirely on the device?</div>
<div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Scaling this to large clusters, replace Xeon Phi by compute node, PCI-Express by GB-Ethernet/Infiniband as well as change the timescale a bit and you're probably not too far off...<div class="im">
</div></blockquote></div><br><br></div></div>