<html><body>
<DIV> </DIV>
<DIV>Gilad Shainer wrote:</DIV>
<DIV> <BR>> With more cores on a single node, the IB benefits are seen in much lower number <BR>> of nodes. I am testing some applications on a new cluster that I have (dual <BR>> sockets quad core Barcelona), and my first results are with Fluent new <BR>> benchmarks. I will have the numbers posted soon, so you all can take a look. For <BR>> 2 nodes, IB shows an average of 15-20% higher performance then GigE, and this <BR>> gap gets bigger with cluster size. At 4 nodes the difference was 40-50%. Even <BR>> more important, 3 nodes results with IB were higher then 8 nodes with GigE, and <BR>> GigE stop scaling after 3-4 nodes (performance numbers were flat after 3-4 <BR>> nodes). </DIV>
<DIV> </DIV>
<DIV>Interesting. I am guessing this is with ConnectX adapters and PCIe Gen2. When does</DIV>
<DIV>the InfiniBand technology used in this test flatten? How about the less expensive SDR</DIV>
<DIV>technology referred to in Jeff's article? Can you provide the curves up to 8 nodes for</DIV>
<DIV>both IB and GE in this test? </DIV>
<DIV> </DIV>
<DIV>A quick analysis, limited to this Fluent test, suggests that 4 nodes plus ConnectX </DIV>
<DIV>technology roughly equals 8 nodes plus on-board GE technology. Can you provide</DIV>
<DIV>the system per node price differences? Then we can rougly determine the cost </DIV>
<DIV>benefit relationship.</DIV>
<DIV> </DIV>
<DIV>Regards,</DIV>
<DIV> </DIV>
<DIV>rbw</DIV>
<DIV> </DIV>
<DIV>
<DIV class=signature id=signature>--<BR><BR>"Making predictions is hard, especially about the future." <BR><BR>Niels Bohr <BR><BR>-- <BR><BR>Richard Walsh <BR>Thrashing River Consulting-- <BR>5605 Alameda St. <BR>Shoreview, MN 55126 <BR><BR>Phone #: 612-382-4620</DIV></DIV>
!DSPAM:47a219a4159828298414181!
</body></html>