In this FAQ thread by ThrillKillKult, he/she shows some of the measurements made for power usage with UD on and off. A value of 14 watts was arrived at, but this is without specifying what type of CPU/system, and what CPU throttle UD was being run at. I suspect it was not a contemporary CPU, especially not a dual-core.

I decided to look into this as I felt that 14 watts seemed a little low, especially with todays CPU's. I've only tested this on two systems so far, both very similar and I will test this when I can get my hands on an Intel dual-core system. One system is an Intel P4 2.8Ghz HT and the other is a P4 3.0Ghz HT, both desktop systems. The individual configurations are irrelevant because we are only measuring the increase in power usage from the quiescent state. What's important here is these are not dual-core machines, thus the difference in measurements between UD and BOINC should be minimal as UD uses almost the entire core.

I ran UD through the entire throttle scale, 10-100%, and measured the raw current being used by the overall system, less its quiescent current. UD is nice in that it uses a very small time slice so the apparent CPU and power utilization follows the throttle setting. The measured line voltage was 117 Vac. Here's the rough numbers:

10% - ~ usage is .15 amp (~18 watts)
20% - ~ usage is .16 amp (~19 watts)
30% - ~ usage is .18 amp (~21 watts)
40% - ~ usage is .20 amp (~23 watts)
50% - ~ usage is .22 amp (~26 watts)
60% - ~ usage is .25 amp (~29 watts)
70% - ~ usage is .28 amp (~33 watts)
80% - ~ usage is .30 amp (~35 watts)
90% - ~ usage is .34 amp (~40 watts)
100% - ~ usage is .40 amp (~47 watts)

BOINC was also tested, but it appears to use a very large time slice (about 1 second) so it simply cycles the CPU between full on and idle. With a setting of 10%, you see the CPU full on for about 1 second, and off for ~8-9 seconds. A setting of 90% is the opposite. At 100%, BOINC was using ~.47 amps (~55 watts), a bit higher than UD due to BOINC running two work units and using the HT portion of the CPU. Keeping the UD monitoring window on-screen used up to an extra .1 amp, but the BOINC manager doesn't appear to suffer from this problem.

The conclusion I came to is 14 watts is a very conservative number for today's PC's. With UD at its default throttle of 60%, the approx power usage is 30 watts, and I would suspect it would be similar for BOINC on a single-core CPU.

I'm not presenting this to be critical, just to present up-to-date facts. In all likelihood the 14 watts was for an older CPU, one that would take much longer per work unit. I suspect we are simply getting faster results for the same amount of power used (watts * work unit completion time). It would be an interesting experiment to record the work unit completion time for various CPU's and the power usage to see if my conjecture is close to correct.

(Jan 5/07: I edited the amps/watts values with more accurate measurements as I have borrowed a digital power-monitoring device used for measuring household appliance, and realizing my analog meter I previously used was out of adjustment.)