On Mar 8, 2012, at 11:06 AM, Lamar Owen <lowen at pari.edu> wrote: > On Thursday, March 08, 2012 10:52:02 AM Les Mikesell wrote: >> Yes, part of the power savings are deceptive - they only kick in when >> the CPUs are idle and your users would be one of the rare cases that >> peg them for long intervals. I think this is getting better in the >> current generation but haven't followed the latest changes. > > In scientific computing, there is no such thing as 'enough cores' and if 3 48 core servers physically fit in the space of three older 6 or 8 core servers, then the users will want to fill that space and get 3 more 48 core servers, and so your power density has doubled. So the '150%' power increase is (if I'm reading Mark correctly) per *rack unit* not per core. And, again, in this space you don't get any savings in power, since this sort of computing eats cores for breakfast. And virtualization to save power will not address this type of user's need. > > I live in the same sort of world, just on a smaller scale, and my biggest power consumer is storage, not compute, but I thoroughly understand Mark's points. So, get more power and UPS. The specs are published, so power consumption shouldn't be a "surprise". -Ross