It used to be just the data center folks that really dug into power consumption and trying to optimize power efficiencies. But the newer machines are eating enough power that even smaller development labs should start taking a serious look when they go to buy hardware – and many are.
I’ve been working around those issues myself for the past several weeks, so I was really pleased to read that SPEC is coming out with a standard benchmark for measuring compute performance to power consumption. SPEC has a formal press release on it as well as details available to the methodology. I’d been doing a crude variation on the theme: Watts/GHz, but I’m well aware that in using that as a measure I’m making a number of hasty generalizations.
One of the most frustrating components of doing this research is actually finding out how much power something is consuming. I’m looking forward to comparing desktops and laptops to the server estimations I did a little earlier. My hasty estimations were mostly done around HP server hardware, using their specs and a downloadable power consumption estimation tool.
What I found was the DL385 series was running around 42 Watts/GHz. The DL585 being slightly more efficient at 41 to 39 Watts/GHz (depending on how you load out the memory). Looking at blades with their shared infrastructure, and you get some even better numbers: the BL465 (equivilant to a DL385 kind of server) ran at 29 Watts/GHz – the best efficiency estimate I got out of the lot, and the BL685c running around 36 Watts/GHz.
I’ve borrowed a power meter – so now it’s on to measuring some actual consumption numbers…