The "Undervolting" Guide

[matmat07]

I lowered all my FID and all but one are on the lowest possible voltage. From 6x to 11x they are all at .9500 V. Would it make any power difference if I only used 12x and 11x?

[unclewebb]

At idle, power consumption hardly varies on Intel CPUs as you adjust your multiplier up or down. They are mostly controlled by voltage. At full load, power consumption varies linearly with MHz so a lower multi will reduce power consumption by a similar percentage.

[unclewebb]

I played around with dual IDA mode some more today and found some limitations like being forced to use the CPU's default IDA VID so this trick isn't going to be for everyone.

I did some more testing and discovered that I was wrong about this. You will be able to use whatever VID you like when Dual IDA mode is enabled.

The only negative so far is that not all laptops support the ability to change SpeedStep on the fly. You need that option in the bios or you need to have the SpeedStep lock bit turned off in order for this trick to work.

[JTravers]

unclewebb, is there a thread dedicated to ThrottleStop somewhere, so we can keep track of its development?

Thanks!

[unclewebb]

ThrottleStop is a multi-purpose tool that it doesn't have a dedicated thread yet.

I did start a new thread in the Software section today where I've explained how to enable Dual IDA mode for users that have a bios that supports this new feature. You need to be able to toggle SpeedStep (EIST) but not all bios versions support this. Luckily my Dell D830 does.

How to Enable Intel Dynamic Acceleration (IDA) on Both Cores of a Core 2 Duo

Give that new thread a bump or two so it doesn't get lost.

I did notice when testing the 45nm T8100 that FID and VID and C-States didn't make any significant difference to power consumption at idle. I think these CPUs idle down to a low power state no matter what you have them set to. If anyone else has a Kill-a-Watt meter or similar, hopefully they can post their own results.

[OMGSnarf]

I've upgraded an Acer 5570z to a T7200, and must have gotten a good sample on eBay - I'm running .9875V on 12x and 11x-6x are stable at .95V. Lowered load temp at 12x multiplier from 87C to 73C. I don't think I applied the AS5 right, might need to disassemble this little guy and try again...whoops.

[matmat07]

At idle, power consumption hardly varies on Intel CPUs as you adjust your multiplier up or down. They are mostly controlled by voltage. At full load, power consumption varies linearly with MHz so a lower multi will reduce power consumption by a similar percentage.

I'm not sure of your answer. I'll to make a question easier to answer. Will a CPU running at 2.0GHz with .95V have the same power consumption than the same CPU running at 1.5GHz with again .95V. Same voltage, but different multiplier.

[unclewebb]

At full load when the VID voltage is the same for both tests, a higher multiplier and higher GHz will consume more power.

At idle, when the VID is the same for both tests, a higher multiplier and higher GHz does not significantly change power consumption. The difference is less than 1 watt and is difficult to accurately measure with a consumer grade Kill-a-Watt meter.

[matmat07]

Thanks, now I understand correctly

[unclewebb]

When testing a T8100, I get zero power consumption difference at idle whether the VID is as high as it can go or as low as it can go and same with the multiplier. C-States enabled or disabled also makes no difference.

The Kill-a-Watt shows a constant 23 watts (0.20 amps) regardless of any changes including whether Minimum processor state is set to 5% or 100%.

I tested a T7300 before and went from high to low with VID and FID and the difference was easy to measure. The same KAW meter went from 20 watts to 27 watts during this test so adjusting your settings correctly would result in a 26% decrease in power consumption. The HP T7300 used Intel graphics so maybe that's why it was a little more efficient.

On a 45nm T8100, I'm seeing zero difference at idle. I can even enable Dual IDA mode and once again, at idle, there is no difference.

Intel must have changed something when they switched to 45nm because regardless of any user settings, these CPUs seem to be in a low power state at idle. It would be interesting to see some more user testing of this.

At full load, lowering the VID definitely makes a difference so it is well worth doing that. I'm going to go do some Prime testing for comparison.

Edit: On a T7300, lowering the VID from 1.200 to 1.000 volts while running Prime 95 Small FFTs reduced power consumption from 56 watts to 42 watts. Big difference.

On the T8100, IDA VID is 1.200 but when running two threads of Prime and with IDA disabled, that drops to 1.1375 volts. Full P95x2 load resulted in 47 watts power consumption and dropping the VID down to 1.000 while running this test only dropped power consumption down to 41 watts. With 45nm, the savings aren't nearly as significant. I'm not sure how stable this CPU is at a VID of 1.000 so the real savings might even be less.

For comparison, with the VID steady at 1.1375 while running P95x2, I dropped the multi to 6.0 and power consumption went down to 39 watts at the wall.

Edit #2: I rebooted and switched to Dual IDA mode. At idle, power consumption was exactly the same as before. At full load with the VID at 1.200 and the multiplier at 11.5 on both cores, power consumption was 51 watts. By lowering VID to 1.050 volts, power consumption dropped to 43 watts while running P95x2 Small FFTs. I tried this test at the default VID of 1.1375 and it was at 47 watts, exactly the same as before. The increase in the multiplier made no significant difference while the CPU was fully loaded as long as VID was the same.