One of the more hopeful signs of an increase in the awareness of energy efficiency in the computing field is that vendors are starting to fight about the results of benchmarking studies.

When Intel and AMD, the two major providers of processing power to the computers of the world, find it worth having an argument about whose servers use less power, it shows that the issue matters, even if the primary goal is to save money rather than save the planet.

An interesting report on Ars Technica unpacks some of the details of the comparison between AMD’s Opteron processors and Intel’s Xeon.

It starts by noting that ‘when power is measured at the wall socket instead of at the processor socket, [AMD] Opteron systems can give better performance per watt than [Intel] Woodcrest systems’, implying that it is not enough to measure processor performance and that a whole system approach is needed.

Author Jon Stokes then points out that different types of memory chip make a big impact on overall power consumption under varying load, with Intel using Intel uses fully-buffered ‘FB-DIMM’ chips while the Opteron uses vanilla DDR2 SDRAM – it stands for ‘double-data-rate two synchronous dynamic random access memory’ and is both simpler and less energy-intensive at low usage rates.

As a result the Opteron systems consume less power overall when under relatively light loads, but the Xeon systems scale better.

The complexity of the analysis, and the continuing uncertainty over which is best shows that we still lack the analytic framework needed to make sensible decisions about energy efficiency in these complex systems. Measuring the power consumption of a fridge is easy, but calculating the best configuration for a server farm remains a real challenge.

The report is at http://arstechnica.com/news.ars/post/200...