On Mon, 30 Oct 2006 at 4:24pm, Feizhou wrote
Do you have any more details on said stability problems (e.g. hardware specs, etc)? I've been very impressed with the Woodcrest Xeon tester I've got -- in fact, it bests my Opterons in almost every benchmark I throw at it, and it's been rock solid. I'm leaning pretty hard in that direction for my next cluster upgrade.
The information about the US Government throwing the Woodcrest solution out was 'reported' by the Inquirer. :D Perhaps IBM boxes...they also posted another interesting bit about Woodcrest and Intel 965 motherboards having problems with raid5...kind of like how there is bad mixing if you use ext3 + 3ware in RAID5 mode maybe. I am afraid that is all I can give you on this.
Ah, yes, the Inquirer. Is there enough salt in the world?
The Korean portal one can be found below but it is entirely in korean.
http://www.inews24.com/php/news_view.php?g_menu=020200&g_serial=225735
Something about overheating and FB-DIMM problems which was only encountered after the thing went live.
Hrm. Sounds like an implementation problem.
How close to live are your benchmarks?
This is what I've got so far http://www.duke.edu/~jlb17/optxeon.pdf. Yes, the Xeons have a slight clockspeed advantage, but not near enough to account for the performance increase. The thermal simulation results surprised me -- I really expected the Opterons to fare better there, given how memory intensive they are. Only the heart phantom sim showed better performance on Opteron due to the onboard memory controller.