On Mon, 2006-02-06 at 10:19 -0500, Chris Dagdigian wrote: > I'm sure that many of you also got the marketing email today entitled > "SGI Altix outperforms AMD Dual-core Opteron > Systems". > > Have not read the paper yet, just passing along the actual URL to > the paper ... > > Skipping the intro and setup pages, the actual PDF is here: It's a really nice paper. I have one tiny little observation. The L3 cache of the 6MB Madison has always made a huge difference. This is what I see as really being measured in these tests: Opteron: 8 CPUs (16 compute cores) sharing 8x1MB L3 = 8MB Itanium: 16 CPUs (16 compute cores) sharing 16x6MB L3 = 96MB I'm not sure apples were compared with apples here, but I guess I'm just picking at hairs. However, non withstanding all this SGI's back plane architecture is still one of the smartest designs out there for mix and match NUMA computing. The paper was a good read thanks for sending it round Chris. Best, J.