[Beowulf] MPI2007 out - strange pop2 results?

Greg Lindahl lindahl at pbm.com
Fri Jul 20 13:40:37 PDT 2007


On Fri, Jul 20, 2007 at 10:43:10AM -0700, Gilad Shainer wrote:

> You had previously compared Mellanox based products on dual single-core
> machines to the "InfiniPath" adapter on dual dual-core machines and
> claim that with InfiniPath there are more Gflops.... This latest release
> follow the same lines...

Strong words from a guy whose only published performance whitepaper
compares Fluent performance using very different versions of
Fluent. (No, they aren't comparable.)

If you're referring to one of my old whitpapers, if I compared
single-core to dual I made it clear, and there was no direct data to
compare to. As you are aware, there's very little published
performance data for Mellanox adaptors, but I'm sure you'll work hard
at making such info available. MPI2007 is a great opportunity to
compare performance: I encourage you to give it your best shot.

Also, your use of "you" is overly wide, let me give you a guide for
telling the difference betweeen Kevin Ball and Greg Lindahl:

Kevin				Greg

diplomatic			rude
physicist			astronomer
short hair			long hair
ballroom dance			renaisssance dance
thin				middle-age spread
tall				claims he's "average height"
too young to be cynical		"you kids! when I was YOUR age..."

> Unlike QLogic InfiniPath adapters, Mellanox provide different InfiniBand
> HCA silicon and adapters.

Er, there are 2 kinds of InfiniPath silicon, thanks for noticing.

I'll be looking forward to seeing lots of MPI2007 benchmark results from
you, Gilad.

-- greg
(used to work for PathScale, apparently still vehement about it. :-)



More information about the Beowulf mailing list