Build Your Own Total System Power Analyzer

Overall Score

The first test simply records the power profile during the boot process to system idle. The boot drive was configured for dual boot with a 30 second default timeout which would then begin the boot process on the default OS, which allowed a non-loaded, non-speed step situation to observe baseline signal to noise.

Boot up all channels

Component Level

A few discussion points regarding the bootup sequence is in order. The first feature of interest is the power profile for the CPU and +12 VDC Aux (hard drives and optical drives). The CPU ramps about 60 W briefly, then plateaus around 80 W with some random structure until it flatlines around 65W for 30 seconds. This is typical during boot up sequences as a) the BIOS and initialization must rectify the VID and set the VRM appropriately and b) the CPU begins executing BIOS code. The 30 second flat line on the CPU is a true idle period before actual OS boot up as this system was setup to default to a dual boot menu with a 30 second delay before selecting the default OS. Second point of interest is the magnitude of the +12 VDC Aux power draw during the intial boot up sequence, spiking as high as 50 W or so (this is for two mechanical hard drives and one optical drive). Not completely unexpected, as the platters must spin up to nominal speed (7200 RPM). The torque needed to accelerate the drives to speed is substantial and draws significant amounts of power.

The second feature is the apparent ‘noise’ in the profile, fortunately, this is where the 30 second pause during the dual boot menu is of use. The random spikes well into the boot up are not noise, but real power fluctuations as the system executes and loads the OS. Zooming in on the boot menu period we can ascertain to some degree the extent of random noise in the measurement. From 105 to 120 seconds within the boot sequence, the total CPU power draw is constant, averaging around 64.7 W, and a standard deviation of 0.1 W. While this is not a perfect noise analysis, we can use this as an estimate of how accurately we can know the power at any given time using this type of device. Within 99% of the distribution, we will know the actual power to within roughly 0.6 W which is more than suitable for these type of measurements.



  • Facebook
  • Twitter
  • Myspace
  • Google Buzz
  • Reddit
  • Stumnleupon
  • Delicious
  • Digg
  • Technorati

Leave A Response