Build Your Own Total System Power Analyzer

Overall Score

Power in personal computing platforms, as well as high performance servers, has become a growing concern both in terms of consumption as well as dissipation. Power is also the most misunderstood, misrepresented measure of computer performance around the web. Much of this chaotic mess hinges around how it is measured and reported. Very few reviewers take a measurement with comparisons at the component level that reveal anything meaningful and, as most often is the case, reported power from one site contradicts the power consumption reported at another. This is derived from the fact that most power measurements are convoluted with other components in the system, as well as inefficiencies in the power supply, while other contributing factors fall squarely on the lack of any standard for extracting power measurements that can be compared at a component or even a system level.

Spec’s power benchmark has come the closest to establishing some semblance of an industry method which would make for appropriate comparisons, but the approach still lacks in terms of universal coverage or specifics to particular components. It also has its own set of limitations and problems. For one, the SPEC power_ssj2008 benchmark measures power at the wall socket, and is a platform power metric encompassing the CPU, MB, memory, and PSU inefficiencies. Furthermore, the SPEC benchmark focuses on one workload, graduated at several different loadings and averaged over a linear progression of those levels, which in turn does not represent real world work loads nor does it expand the breadth of programs that can create completely different power profiles. Real World Tech has published a very comprehensive analysis of the SPEC power benchmark here, it is highly informative and suggested reading.

Compounding this mass hysteria of discombobulated nonsense are the ramblings eminating from the PR departments from many companies, each claiming the most power effiecient products and developing new terms and metrics that lend credence to their claims. Some place weight on dissipation, others on consumption or the two concepts are interchanged without reconcilliation. Neither does it help that definitions are vague, specs are ill-defined or ambiguous, and marketing departments choose to keep the entire topic obfuscated– most likely by design. Marketing’s motto here would seem to be ‘what the end user can’t understand won’t hurt us’.

Power as the end user, or enthusiast, should comprehend it falls into two basic concepts. First, how much will it hit the pocket book — consumption — and second, what type of cooling (and where should one cool the most) — dissipation. These two ideas are intertwined, however distinctly different. Intel or AMD, unlike many other advanced device manufacturers, make thermal dissipation power (TDP) a major point on their products; however, this number is often confused with actual consumption treating TDP like the rating of an incandecent light bulb, i.e. a 95 W TDP is mistakenly assumed to mean the processor consumes 95 W of power. AMD has gone to great lengths, for example, to differentiate TDP from actual consumption by establishing a different concept called ACP (average consumed power), which is distinctly different than TDP in both usage and meaning.

What would be useful, although difficult, would be a method for accurately tracking power (real time) as a result of all major components either within itself or in addition to a system level (at the wall) power study. Too many reviews around the web utilize a standard consumer based power meter (typically a Kill-O-Watt meter) plugged into a wall and simply watch the meter for peak power. This is great for the ‘what will it hit me in the pocket book’ crowd (so long as you can trust that the user who records the data is objectively observing the correct result). The power at the wall approach, however, is poor when analyzing performance of individual components and is not at all informative in helping end users, like you and me, in making decisions when piecing together a system with power as the major concern.

Ultimately, everyone would be best served if, in addition to reporting total power at the wall, power consumed by CPU, GPU, and motherboard were analyzed as well. This way both the total system power, as well as where the power is parititioned, can be rationalized and systematic errors eliminated, at least those introduced by the PSU itself. What you will read about here, today, is a nifty method for not only measuring power on all the voltage rails but also how to build this system economically. The article will conclude with some simple experimental results that looks at different software commonly used to produce a load, and show that different results can be obtained simply by the software one may use.

SHARE THIS POST

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

Leave A Response