E8400 power management, not worth the trouble?

[amyhughes]

From my lowest-stable-voltage tests of my E8400:

333x6 = 2.0 GHz (min multiplier), 0.8125 V set in BIOS
333x9 = 3.0 GHz (max multiplier), 1.0125 V set in BIOS

The difference in total system power draw from the wall between these two states at idle is 4W. 71W vs 75W. The best power management strategy in the world couldn't do better than 4W at idle. Intermediate load states would save even less. C1E doesn't save anything at all (75W idle throttled to 2.0 GHz).

With overclocking you can save a little more:

400x6 = 2.4 GHz, 0.8875 V
400x9 = 3.6 GHz, 1.1625 V

7W difference at idle. 78W vs 85W. C1E manages 2W savings (83W idle throttled to 2.4 GHz).

The lowest voltage I can post with is 0.775V, at 300x6=1.8 GHz. That draws 70W.

prime95 load at 333x9=3.0 GHz, 1.0125V, draws 104W.
prime95 load at 400x9=3.6 GHz, 1.1625V, draws 128W.

[Modo]

While the power savings might not be impressive, undervolting is good for keeping the system silent. It shaved 5C of the idle temperatures on my CPU (E6550), and a good 10C at load. That's enough to keep the fans at low speeds with moderate load (in games, which is the highest load I actually use, except when stress-testing on purpose).

[amyhughes]

While the power savings might not be impressive, undervolting is good for keeping the system silent. It shaved 5C of the idle temperatures on my CPU (E6550), and a good 10C at load.

The numbers I listed *are* undervolted. My point is that the savings you can get with fancy power management (e.g. RMClock) with an already-undervolted E8400 is slight. You'll get 0W and 0C savings at load because the voltage I've given is the lowest stable at load. The maximum savings to be had is at idle, and it's only 4W.

[BillyBuerger]

That's very interesting. And makes me wonder why Intel won't do 1GHz idle states? Is there a problem with a 3x multiplier? Or are they worried about people having issues with it not clocking up and bitching about performance. I've found links about AMD's Cool'n'Quiet where people were complaining because it by default wouldn't clock up until over 50% of the CPU was used. If you have a dual-core CPU and a single CPU process using 100% of one cpu, you're under the threshold and are only running the single process at half speed. With a higher idle clock, this isn't as noticeable. Although you would think this could be fixed in the driver. RMClock let's you work around this issue too.

I still love my Pentium M 1.6GHz. Idle at 600Mhz and 0.7V. That from a Celeron M 1.3GHz locked at 1.3V (or something) was a huge difference in my old laptop. And it's doing pretty good in my HTPC although Blu-ray is definitely pushing it.

[amyhughes]

makes me wonder why Intel won't do 1GHz idle states? Is there a problem with a 3x multiplier?

Maybe there's no power to be saved. Graphs I've seen of measured power usage of the processor alone show 45nm chips using low single digit wattage at idle. I can't even try a 3x multiplier, but at 6x the lowest wattage I've gotten is 70 (full-system), and at 2 GHz I get 71W. I suspect if I tried 100 MHz FSB I'd get no less than 69W. I'd try it but at the low end of voltage it's easy to get it so it doesn't post, and I'm kinda tired of clearing CMOS

[XS Janus]

0.8125 V set in BIOS
Is a great voltage.
I have 8200 and mine won't get below 0.975v in bios.

I'll have to find my measurements of consumption, but I remember them being similar to your deltas.

[smilingcrow]

Intel are stuck on a minimum multiplier of 6 for some reason even with the mobile CPUs. The current mobile platform does reduce the FSB from 800 to 400 at idle giving a frequency of 600MHz. I have a C2D laptop that supports this feature but haven’t measured the power saving yet. RMClock now handles this down clocking.

[yuu]

multiplayer of 6 worsens l2 cache latency. probably it isn't that bad.