Page 1 of 1

Undervolting Core2Duo

Posted: Sun Nov 05, 2006 9:27 pm
by GHz
Hey,

I'm surprised I didn't see a topic on this, so I thought I'd start one. If you have a Core 2 Duo CPU, please post your undervolting experiences! I'm done testing the overclocking limits, so now I'm going in the other direction.

Specifications:
=================================
Intel Core 2 Duo E6600 2.4 GHz
Gigabyte GA-P965-DS3, F7 BIOS
2 x 1 GB Patriot PDC22G5300LLK DDR2-667
ATI Radeon X1900XT 512 MB
SeaSonic S12-500, Rev. 2
BlueGears b-Enspirer 7.1

Test Equipment:
=================================
P3 International KILL-A-WATT AC power meter

Results*:
=================================
3.5 GHz (389x9), 1.487v, 157W idle, 214W Prime95, 239W TAT
3.4 GHz (378x9), 1.425v, 147W idle, 200W Prime95, 220W TAT, 14.9s SuperPI
3.3 GHz (367x9), 1.375v, 141W idle, 188W Prime95, 205W TAT, 15.5s SuperPI
3.0 GHz (334x9), 1.250v, 126W idle, 163W Prime95, 173W TAT, 17s SuperPI
2.4 GHz (266x9), 1.325v, 117W idle, 163W Prime95, 170W TAT, 21.1s SuperPI**
2.4 GHz (266x9), 1.1625v, 117W idle, 145W Prime95, 146W TAT, 21.1s SuperPI**
2.4 GHz (266x9), 1.025v, 109W idle, 133W Prime95, 134W TAT, 21.1s SuperPI**

* EIST disabled, C1E halt enabled for all tests. CPU normal vcore reads as 1.325 volts. Memory kept at or above DDR2-667 (4-4-4-12) using 2.0-2.5x multiplier. All results were verified stable by passing 12 or more hours of Prime95 (actually, SP2004 with custom settings to max out memory usage) AND 12 hours of 3DMARK06 (racked up 100 KWh this month on computer alone to prove it!).

** This result is using stock settings, except for 2.5x memory multiplier to reach DDR2-667. EIST and C1E were enabled.

I'll post more results as I have them.

EDIT: 11/6/06 - More data: stock results, SuperPi
EDIT: 11/12/06 - Results at minimum EIST voltage (1.1625v)
EDIT: 11/13/06 - Results at 1.025v

Posted: Mon Nov 06, 2006 1:38 am
by Sendorm
Thanks for the results.

Can you also add the watt results for default voltage and default frequency.

Also, can you add the systems performance value for each result (for instance superpi 2m results or 3d mark cpu results) so that we can create a performance/watt for each case. To see if undervolting really helps or not. But my guess is the p/w parameter would be quite close for each case.

Posted: Mon Nov 06, 2006 5:51 am
by kike_1974
I have one question about C2D undervolting that maybe you can ask. Which is the minimun vcore that you can select (using crystalcpuid or rmclock or similar)?
I know that E3 and E6 revisions of Athlons have a 1,1V lock for Vcore, and I would like to know what is the limit of Core2Duos.

Posted: Mon Nov 06, 2006 6:10 am
by Schlotkins
This is great work - I'd also love to see some results for the default settings.

Cheers,
Chris

Posted: Mon Nov 06, 2006 8:57 am
by GHz
Stock & SuperPi results added. It looks like the overclocked/undervolted settings at 3.0 GHz/1.25v make it run at the same power efficiency as stock at 2.4 GHz/1.325v, except for idle power using EIST. For example, at 2.4/1.325v with EIST disabled, power consumption was 128W compared to 3.0/1.25v at 126W. I'll have to tinker with EIST to see if I can keep it enabled while overclocked/undervolted.

EDIT: Looks like EIST works at 3.0 GHz if I keep voltage at AUTO in the BIOS. This drops idle power to 121W. I can manupulate the settings using RMClock. Voltage range using EIST/RMClock is 1.162-1.325v versus 0.5125-2.0v in the BIOS. I could probably run stock 2.4 GHz at less than the minimum EIST voltage by setting the vcore manually in the BIOS :D

Posted: Mon Nov 06, 2006 6:14 pm
by Schlotkins
Impressive results on that chip. I wonder if most are like that.
Looks like 3.0ghz is the best performance/watt.

Chris

Posted: Mon Nov 06, 2006 6:16 pm
by GHz
I can't wait for SPCR to do another CPU round-up ;)

Posted: Mon Nov 06, 2006 10:07 pm
by Sendorm
Ok let's see, the performance per watt parameter for load cases.

Performance parameter is (1/superpi), so to get the performance per watt value we shall calculate 1/(superpi*watt). I will be using the prime95 wattage, also note that bigger is better and finally the results are multiplied by 1000 for your eye candy.

3.4 GHz (378x9), 1.425v, 1/(200W*14.9s) = 0.3355
3.3 GHz (367x9), 1.375v, 1/(188W*15.5s) = 0.3431
3.0 GHz (334x9), 1.250v, 1/(163W*17s) = 0.3608
2.4 GHz (266x9), 1.325v, 1/(163W*21.1s) = 0.2907

So these numbers prove that undervolting does work and similary it shows that the stock case is the worse. Thanks GHZ.

Posted: Tue Nov 07, 2006 4:03 am
by PsyDuck
What's the lowest working voltage at stock speed?

Posted: Tue Nov 07, 2006 6:01 am
by kike_1974
And what is the lowest voltage at lowest possible speed? (Interesting for lowest consumption at idle)

Re: Undervolting Core2Duo

Posted: Tue Nov 07, 2006 6:17 am
by smilingcrow
GHz wrote:I'm surprised I didn't see a topic on this..
There are quite a few topics on this, I should know I've posted quite a few myself. ;)
If you do a search by my user name for topics in this sub forum, you will see which ones I've started.

Re: Undervolting Core2Duo

Posted: Tue Nov 07, 2006 6:40 am
by Mats
GHz wrote:3.5 GHz (389x9), 1.487v, 157W idle, 214W Prime95, 239W TAT
Call me no°b or whatever, but what does TAT mean? I guess it's not Tiffani-Amber Thiessen.

Posted: Tue Nov 07, 2006 7:33 am
by gb115b
i thought they automagically under-clocked to 1.6ghz on idle? or am i thinkign of something different....

Re: Undervolting Core2Duo

Posted: Tue Nov 07, 2006 7:41 am
by jaganath
Mats wrote:
GHz wrote:3.5 GHz (389x9), 1.487v, 157W idle, 214W Prime95, 239W TAT
Call me no°b or whatever, but what does TAT mean? I guess it's not Tiffani-Amber Thiessen.
:wink: Unfortunately not, it's Intel's Thermal Analysis Tool.

Re: Undervolting Core2Duo

Posted: Tue Nov 07, 2006 12:32 pm
by Mats
jaganath wrote: :wink: Unfortunately not, it's Intel's Thermal Analysis Tool.
Thanks! Well I wondered how she would make the PC hotter, using 25 W more...

Posted: Tue Nov 07, 2006 5:01 pm
by cmthomson
gb115b wrote:i thought they automagically under-clocked to 1.6ghz on idle? or am i thinkign of something different....
By default, C2D CPUs drop to 6x under light CPU load on most motherboards. They also drop to about 1.05V at this speed from the nominal 1.35V.

If the FSB is not overclocked, 6x266 -> 1.6 GHz.

It is possible using RMClock to always run at the lower settings to get really low power consumption. This of course only makes sense with a 6300.

But many C2D owners go the other way: overclocking 6600s to the max, and overvolting to get there. 40-50% overclocking is not unusual. To do this, on most motherboards, the C1E and SpeedStep features have to be disabled to achieve stability. This is because if you're running at the ragged edge of reliability at (say) 9x370 and 1.5V, you'll probably crash or be unable to boot at 6x and 1.05V.

Posted: Tue Nov 07, 2006 5:13 pm
by kike_1974
Ok, so let's assume a not overclocked E6600 at idle (at 6x266=1.6GHz and being not overclocked and not running at the edge of stability, CE1 and speedstep can be enabled). What is the lowest voltage that you can set with RMClock to get the lowest power consumption, and what is the power consumption at that point? I suppose that you can set a lower value than the default 1.05V. Or even more, maybe it is possible to set a lower multiplier if it is not locked (for example at 5x266=1,33GHz).
Could you try this? Thanks :D
I have seen many experimets to reach the upper limits, but haven't seem this other to reach the lower limits.
cmthomson wrote:
gb115b wrote:i thought they automagically under-clocked to 1.6ghz on idle? or am i thinkign of something different....
By default, C2D CPUs drop to 6x under light CPU load on most motherboards. They also drop to about 1.05V at this speed from the nominal 1.35V.

If the FSB is not overclocked, 6x266 -> 1.6 GHz.

It is possible using RMClock to always run at the lower settings to get really low power consumption. This of course only makes sense with a 6300.

But many C2D owners go the other way: overclocking 6600s to the max, and overvolting to get there. 40-50% overclocking is not unusual. To do this, on most motherboards, the C1E and SpeedStep features have to be disabled to achieve stability. This is because if you're running at the ragged edge of reliability at (say) 9x370 and 1.5V, you'll probably crash or be unable to boot at 6x and 1.05V.

Posted: Tue Nov 07, 2006 5:36 pm
by cmthomson
kike_1974 wrote:Ok, so assuming a not overclocked E6600 at idle (at 6x266=1.6GHz), what is the lowest voltage that you can set with RMClock to get the lowest power consumption, and what is the power consumption at that point? I suppose that you can set a lower value than the default 1.05V.
Could you try this? Thanks :D
Sorry, not me. I'm the mad overclocker from hell. :twisted:

Search the forums for smilingcrow, who did some of these experiments.

Power consumption varies according to V*V*F, so lowering the voltage is very significant. I would estimate a 6600 at 6x and 1.05V under full load to be 33W. At idle it would be less than half that.

Posted: Wed Nov 08, 2006 4:24 am
by kike_1974
I found some interesting information about undervolting and overclocking in this post:
My experience on undervolting & overclocking Core 2 Duo
I've seen there that the lowest voltage that it can be used is 1,15V (for E6400 at least) and not 1,05V.
EDIT:
I was kinda blind when I looked for posts about this. There is this post from smilingcrow:
C2D, low power at idle & high O/C with new BIOS
There it shows that the lowest voltage that can be selected with RMClock or CrystalCPUID is 1,05V for E6600 and E6300. Not good news, but it is still better that the 1,1V lock for X2 Athlons.

Posted: Wed Nov 08, 2006 6:35 am
by Mats
I wonder if revision B3 will bring down the power consumption even more.

Re: Undervolting Core2Duo

Posted: Wed Nov 08, 2006 8:34 am
by genericname
GHz wrote:Intel Core 2 Duo E6600 2.4 GHz
Gigabyte GA-P965-DS3, F7 BIOS
Same here, but my BIOS reports a stock voltage of 1.275V. When CPU voltage is set to Auto in BIOS, EasyTune reports even lower voltages at stock speed (2.4GHz). At 3.0GHz, it autoadjusted voltage up to 1.39V, although I'm currently typing this at 3.0GHz and 1.28V. I will try lower voltages.

Posted: Wed Nov 08, 2006 8:48 am
by jaganath
I'm currently typing this at 3.0GHz and 1.28V.
Incredible, 3G's at basically stock volts; the C2D is an overclocker's dream!

Posted: Fri Nov 10, 2006 8:25 pm
by GHz
Sorry for letting this thread sit idle. I don't like to report my readings until I'm sure they are 100% stable, but I haven't been able to do the full gamut of tests because I've been busy. Also, I was having trouble maintaining stability at totally stock settings (i.e. defaults) at low voltage. So I left my OC settings intact and changed only what I needed: FSB and vcore. Here's the part I know some of you are dying to hear:

2.4 GHz @ 1.1625v (min. EIST setting!), 120W idle, 140W Prime95, 147W TAT

These settings passed 6 hours of Prime95/Orthos before I had to cancel early (had to do some work).

Posted: Sat Nov 11, 2006 2:35 am
by Poodle
GHz wrote:Sorry for letting this thread sit idle. I don't like to report my readings until I'm sure they are 100% stable, but I haven't been able to do the full gamut of tests because I've been busy. Also, I was having trouble maintaining stability at totally stock settings (i.e. defaults) at low voltage. So I left my OC settings intact and changed only what I needed: FSB and vcore. Here's the part I know some of you are dying to hear:

2.4 GHz @ 1.1625v (min. EIST setting!), 120W idle, 140W Prime95, 147W TAT

These settings passed 6 hours of Prime95/Orthos before I had to cancel early (had to do some work).

Nicely done! Not bad for a 1900xt system.

Posted: Sun Nov 12, 2006 10:17 am
by GHz
Updated my first post with results using EIST and C1E. Stable for 12 hours of Prime95/Orthos:

2.4 GHz @ 1.1625, 117W idle, 145W Prime95, 146W TAT

I'd love to know what the TDP of this CPU is at this point. Anyone want to take a guess? I'm going to trying dropping voltage down further :D

Posted: Sun Nov 12, 2006 11:41 am
by Mats
GHz wrote:I'd love to know what the TDP of this CPU is at this point. Anyone want to take a guess?
65 x 2400/2667 x 1.1625²/1.3525² = 43 W :P

Posted: Mon Nov 13, 2006 8:39 am
by GHz
OK, it looks like I've found the low-voltage limit. If I go any lower at this speed, I can't pass Prime95:

2.4 GHz @ 1.025v, 109W idle, 133W Prime95, 134 TAT

That comes out to a maximum TDP of 35W. I think we've found our new undervolting champion :D

Re: Undervolting Core2Duo

Posted: Thu Jan 25, 2007 7:02 am
by jojo4u
GHz wrote:Hey,
2.4 GHz (266x9), 1.1625v, 117W idle, 145W Prime95, 146W TAT, 21.1s SuperPI**
2.4 GHz (266x9), 1.025v, 109W idle, 133W Prime95, 134W TAT, 21.1s SuperPI**
** This result is using stock settings, except for 2.5x memory multiplier to reach DDR2-667. EIST and C1E were enabled.
Interesting. With enabled EIST/C1E you nevertheless get lower power draw at 1.025 V. Had you enabled minimal power consumption in system control? I'd really like to see CPU-Z and RMClock monitoring pictures of (EIST/C1E on and EIST off/C1E on)@1.025 V.

Posted: Thu Jan 25, 2007 8:22 am
by Mikael
System:

Antec NeoHE 430W
E6600
Gigabyte GA-965P-DS3
2*1GB Kingmax DDR2-667
GeForce 7900GS @ 520/720
Soundblaster Audigy 4
WD Caviar SE16 250GB
Old NEC CD-ROM

2.4GHz @ 1.0V BIOS / 0.98V CPU-Z

3.0GHz @ 1.17V BIOS / 1.14V CPU-Z

Judging by the voltages, this is without doubt a pretty good CPU. The first config was tested a few hours with Orthos and the second was tested 8+ hours with Orthos and a few 3DMark06 runs. No stability issues.

I don't have all the power measurements at hand now, but I'll try to get back with them. I do know that the first config idled at 101W and that the second was at 156W while running Orthos. The exact same system, but with an Nforce 550 board and an X2 3800+ AM2 @ 2.2GHz and 1.2V required 108W at idle and 144 under load. Pretty impressive that the vastly superior 3GHz C2D only needs 12W more at the wall socket.