My iPod is playing “Everyday is Like Sunday” by Morrissey. I am currently somewhere between Russia and Alaska on a plane home from 2 weeks of vacation in Japan. Woke up this morning about 6, and at this rate, I won’t sleep for about 36 hours.  Every day may be like Sunday, but this Sunday is like part of Monday too.

With lots of time to kill, I have been thinking about designing for power efficiency.  The latest generation of server processors are definitely designed to focus on the energy efficiency needs of the data center, but how do companies go about doing this? Is it all the same?

I took a look at a recent article at Tech Report.com that highlights, amongst other things, some power efficiency calculations they performed on both our latest six-core AMD Opteron^TM processor and our competitor’s latest release. These are Tech Report’s test  results – I haven’t had the time or the opportunity to recreate the tests in my own lab, but I trust their methodology, and believe these results are reasonable and accurate.

One thing to note is that in all of the numbers below, we are looking at our six cores versus our competitor’s four cores, so from a core count, when it comes to power, we have a “hand tied behind our backs” with 50% more cores drawing power. Plus, they configured the competitor’s system with 6 DIMMs vs. our 8 DIMMs.

One of the most striking points, according to Tech Report, is that our six core platform and their four core platform both idle at approximately the same level – our 154W to their 153W. 2 extra cores for only 1 watt, what a great deal.

However, the platforms diverge when you add the slightest load. With only a 2% load, we moved from 154W to 158W, not a large delta at all.  Our competitor, however, jumped from 153W to 186W. That is a pretty dramatic jump in power consumption.  If you consider that throughout the course of a day (even a long one like today), the processors will jump back and forth between idle and light loads quite often. Having a low idle power on a server is like having low gas consumption on a car in idle.  Eventually you are going to want to get somewhere; people don’t buy cars to sit on their driveways.

Based on my interactions with customers, I’ve found that most IT managers believe that their average CPU utilization is ~15-20%.  This means the processors are probably spending a lot of time over the course of the day going in and out of idle.  So these small jumps in utilization could mean big jumps in power consumption, all depending on whether you make the right processor choice of course.

Peak power is another story.  Our peak power in the Tech Report article was 278W, quite respectable for a system with 12 total cores. Our competitor’s Nehalem-based system pulled 330W at the wall when in peak load.  52 watts is a lot of power, and if you are running virtualization or an HPC cluster, where high utilization is the profile, then you are really limiting yourself by not choosing these AMD Opteron processors.

Data center floor space is expensive, probably the most expensive area in all of your facilities. Those extra 52 watts add up quickly.  With a 1200KVA power budget per rack you could fit 42 1U AMD Opteron processor-based platforms, or only 36 competing 1U servers. That means potentially wasting almost 15% of the space in each rack.  Meaning more racks, and more valuable floor space consumed.

It’s not enough to say that you design processors for energy efficiency.  You have to optimize across the whole range of utilization levels.  Only then can you be sure that whether you are dealing with a low level of utilization or a highly utilized server that you are getting the most for your money.  No matter how long your day is.

John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.