Apple dropped me a note in response to my blog post on its Harpertown Xserve and Mac Pro announcement. I attributed the per-socket CPU power draw claim of "80 watts max, 4 watts idle" to Intel. That turns out to be Apple's number, not Intel's.
I'm not much interested in Intel's stated Harpertown per-socket power draw because I can't reproduce Intel's test conditions. Outside Intel's labs, you can't pin down a single component's true power draw without a well-equipped test bench and a very steady hand.
If you have an Xserve or Mac Pro, you can skip the bench and skip Intel's data sheets as well. Apple builds an uncommon level of instrumentation into Xserve and Mac Pro. OS X Server Leopard's (or Tiger's) Server Monitor reports on component-level power draw and fine-grained regional temperatures in real-time. You can subject Xserve or Mac Pro to varying workloads and track power utilization of CPUs, DIMM sockets and the Intel north bridge independently. It is through this facility that I learned that Intel's north bridge (memory and I/O hub) chip is the least green component in the system.
I was green before green was in, and I am a firm believer that the only place to measure power draw is at the outlet. But chipmakers, and OEMs who ride the shirttails of chipmakers' marketing, compete based on power consumption per CPU socket without providing consumers or product testers the means to validate their claims. At least with Xserve and Mac Pro, I can see for myself. The figures may not be absolute--they can only safely be compared Apples to Apples--but Server Monitor will reveal whether Harpertown's 45 nanometer-ness is directly related to its greenness. With faster front side and memory busses, will cooler CPU sockets matter? As you can tell, I'm eager to find out.