Today, feeding my own curiosity, I had a chance to finally measure the power consumption for several computers under real-life load and results are somewhat different from what I thought they would be.
First of all, the computers in test:
1. Lenovo X61s – Intel L7500 1.6 GHz
2. Lenovo T61 – Intel T7700 2.4 GHz
3. ASUS P5 based PC – Intel Q6600 2.4 GHz
4. XFX Geforce 9300 based PC – Intel E7500 2.93 GHz (used as an HTPC)
5. ABIT BE7 based PC – Intel P4 2.4 GHz (used as a web server for hosting this very blog and some other websites as of 02.26.2010)
The device I used for measuring is blueplanet EM100 electronic energy meter.
With my first round of tests I just measured the average consumption under no load with speedstep on, then I measured the the average consumption with cpu’s maximum speed and no load and then the wattage with maximum speed and maximum load.
It must be noted, that I initially ran into difficulty with enabling speedstep on Q6600, but that was resolved with ASUS P5B’s bios update.
For myself I’m trying to figure out whether it would be a big difference between Q6600 and E7500 in terms of energy consumption, and whether P4 based server should be written off and either Q6600 or E7500 should take it’s place. I should admit, that if I were to buy server hardware from scratch at this point in time, I would probably go with Atom-based solution, since those CPUs have amazingly low power consumption matched to decent performance.
In my preliminary results the x61s shares the first place with HTPC based on E7500. I’m not sure why HTPC consumes that little – whether it has to do with new 45 nm technology or because the power supply that came with the multimedia case is rated at lower wattage than the other two cases (P4′s and Q6600′s). The second place is occupied by T60 – no surprises here. The main surprise is that the 3rd place is pretty much shared between Q6600 and P4, which is interesting, considering that they are pretty much on the opposite ends of the performance scale. In other words, running Q6600 based server will essentially cost me the same in energy costs, while in terms of CPU resources it will provide much more, which makes it more economically viable solution than P4.
Alternatively, it makes a lot of sense to select hardware with a specific purpose in mind, since I’m pretty sure that the motherboard with on-board video will consume less power than motherboard/video combination, which is not required for a server anyway.
I will publish the detailed test results in the next entry on this topic, along with more thoughts on this.
Tags: Green Computing, Hardware, HTPC