Indepth E6400 Power Consumption Analysis
01.21.07 - 09:25pm
I’ve been silent here for nearly a week and you can blame HWBot.org for that. Recently my home forum and team for overclocking, XtremeSystems, lost their 1st place and there was a call put out for anyone and everyone to help take back #1. For those of you not familiar with HWBot, it is essentially a ranking of overclocking records for various benchmark programs that have set the bar for overclocking and stability. This particular calling came out at an opportune time as just a week before I had finished up my singlestage phase cooler and had already been benching my system. So rather than spend my time doing productive things like write, I spent my time running loops of benchmarks watching the same scenes render over and over again. However, while watching those scenes render I did gather a ton of data and what the common person may find most interesting was my power consumption data. So further blubbering aside, here we go.
Late last year I wrote an article complaining about the sudden lack of notice in the computing industry over power consumption. In the Pentium 4 era power consumption and heat were always an issue and this lead to the development of considerably less powerhungry processors. First came the Pentium M Dothans with their sub-3GHz clockspeeds and decent single-core performance but the chip was destined to become a mobile processor. Then came Dothan’s replacement, Yonah and what a surprise Yonah was. Yonah featured two massaged Dothan cores squished together with a unified cache. Unlike the Pentium D, Yonah was actually 3 steps forward and 0 steps backwards. After Yonah came Conroe and what a beast it was. Conroe proved capable of scaling from a lowly 1.86GHz up to 2.93GHz from the factory and capable of 5.5GHz and greater in the right hands. However along with this performance came a terrible cost in the form of heat and power consumption. The increased amounts of core voltage required for stability made this “65Watt” processor into something capable of consuming more than double its rated capacity.
I tried to be as scientific as possible with the tests that I performed but there is still a lot of room for error on half of the data provided. Rather than just show you the power consumption at idle and full load, I decided to throw in the POST and Windows Loading values. While the computer rarely will spend more than a few seconds in these states, I find the data to be most interesting. The idle power numbers were attained after the system had booted into Windows and had been idle for roughly 5 minutes. The full load numbers were measured after the system had clocked 10 minutes through a dual SuperPi 32M run. I chose SuperPi32M as it completely loads both cores on the processor and works over the memory subsystem rather well while leaving the graphics system at idle. I tried to keep the graphics system at idle as my video card is capable of rivaling the CPU in regards to power consumption but I couldn’t totally remove it from the equation. I am confident though that the GPU was pulling as much current as 3-4 hard drives would pull so the results can still roughly apply to a server environment.
At stock speeds the power consumption was shockingly low. I was expecting to see the system pulling atleast 200 watts but at full load the system drew a meager 143 watts. Bumping up the FSB and Vcore slightly saw a slight increase in power consumption but still considerably acceptable with 158 watts being drawn. The next jump was considerably, going from 2.66GHz to 3.2GHz and the power consumption jumped to 184 watts. From 3.2 GHz up to 3.8GHz the system gradually drew more power as the Vcore was increased to maintain stability but the results weren’t terribly mindboggling with the system pulling just over 200 watts at 3.8GHz. When you think that just 1 year ago the Pentium Ds were capable of drawing 200 watts by themselves at 3.8GHz, this is rather remarkable. Even at 4GHz the system only pulls 227 watts at full load, rather remarkable for 8GHz of combined power.
Now these results may not apply to all processors as various chips respond to voltage increases differently. My particular chip responds really well to voltage and is capable of scaling from 3.9Ghz to 4.3GHz with only 1.525 volts, a remarkable feat I might add. One note that should be made though, even with the enormous Scythe Ninja I was not able to get 4GHz stable under air-cooling, I had to strap my singlestage to the processor to get 4GHz stable. I do wish I had an E6600 and Q6600 to test but my budget simply doesn’t have room for either of these chips right now. Once I get a quad-core processor I will run an identical battery of tests and see how those chips stack up. Prices should fall considerably in the coming months and I’ll do my best to snag a few more processors to test.
Speak Your Mind