Advantages of undervolting the Intel Core Duo

Introduction

This is a brief experiment aiming to determine the effects of undervolting Intel Core Duo CPUs. As we probably all know by now, it is impossible to undervolt the Core Duo at the lowest multiplier setting, because that voltage value is locked into the hardware. Therefore, undervolting will decrease neither the idle temperature nor the idle power consumption of the CPU, whereas this could be done for the Intel Pentium M. So, we will get no increase in battery life by undervolting for our Core Duo-driven notebooks.

Nevertheless, for those of us who use many times our notebook at the top CPU speed or close to it (e.g., for running numerical simulations, video encoding, etc.) it is interesting to determine what are the advantages, if any, of undervolting the CPU at higher multiplier values.

The conclusions of this experiment might apply to the Intel Core 2 Duo, as well.

Experiment description:

CPU type: Intel Core Duo T2400 (1.83GHz)
Notebook type: ASUS V6J

Ran CPU at full blast starting from idle until the CPU temperature stabilized, with three undervolting settings:
E1. Baseline: CPU not undervolted, VID at multiplier 11x = 1.262 V (ran twice)
E2. CPU undervolted 0.1V, VID at multiplier 11x = 1.150 V (ran twice)
E3. CPU undervolted 0.2V, VID at multiplier 11x = 1.050 V (ran once)

Results:



The results are depicted in the figure above. Information relating to setting E1 is colored blue, for E2 green, for E3 magenta. The moments of starting to run the CPU at full blast are marked with blue, green, and magenta vertical lines; those for returning it to idle and 6x multiplier are marked with red vertical lines. The various time and temperature intervals denoted in the figure are all discussed in the next section.

The middle part of the figure between experiments in settings E2 and E3 has been cut out for better readability (there was an interval during which nothing happenned).

Discussion:

Undervolting significantly helps in decreasing the temperature of the CPU at full blast. More specifically, a drop of 0.1V in VID resulted in an 8 Celsius degrees difference between the settling temperatures of E1 and E2 (\Delta T_1).

When the drop in voltage was increased to 0.2V for experiment E3, the difference increased to 10 Celsius degrees (\Delta T_2). So, the dependence of the temperature drop on the voltage is not linear (i.e., cutting off the voltage with the same amount of 0.1 between E2 and E3 only gains 2 more degrees instead of 8 as was the case between E1 and E2). Nevertheless, we can see another interesting thing: the time it took the CPU to settle to the maximum temperature is much larger in experiment E3 than in experiment E2. So although the final gain in temperature is not so significant when the voltage is dropped further, in the transient before the temperature settles the gains are again significant.

Other observations:
- Depending on the starting temperature, the final settling temperature will differ. Specifically, 2 was run twice, once starting from 60C and then from 55C. The final settling temperature was 0.9C larger in the first case than in the second (\delta T).
- The temperature decrease helps in decreasing fan effort and noise. This was obvious on my V6J in E3 when the fan did not need to go to full blast in order to cool the CPU. In E2 it did, as of course in E1.
- As a control engineer, I find it nice to see the nice exponential settling curve of the temperature, which means the actively-cooled CPU's temperature response can be approximated with a first-order dynamical system.


Software used:
MobileMeter was used for monitoring the temperature.
RMClock 2.15 was used to control the CPU in E2 and E3.
Intel Thermal Analysis Tool 2.05 was used to load the CPU to maximum during the experiments.

 

An excellent piece of work with some graphs graphs. It probably belongs better in the hardware forum in order to receive wider acclaim.

I have got similar results for my Samsung notebooks with around a 10°C drop in the maximum temperature and the fans never now go up to their maximum speed. For the T5600 in my Q35 which runs fine at 1.0V at maximum speed, I measured the drop in power consumed, as a result of the undervolting, as around 10W (there are some comments about this near the end of my Q35 review).

The T7200 in my X60plus came with the default voltage of 1.187V at full speed (this has been running happily down at 1.05V max), while the T5600 had a default voltage of 1.25V at full speed. I haven't understood the basis for these pre-set voltages: I would have expected all the C2D CPUs to be set to the same default voltage.

John

 

You can just move it to the hardware forum if you wish, I just posted here since this is my "home", so to speak.

 

How do you do something like this, and am I right to assume there is performance loss in doing it?

 

There is no performance loss. It's simply that (most) individual CPUs work with a lower voltage than the "safe" voltage specified by the designers. Probably you have more sensitivity to things like interference from other components, and below a certain voltage limit stability is lost, but empirically people have observed that running the CPU at 0.2V lower than the specification is usually safe.

And you do it using undervolting software like NHC and RMClock.

 

"Voltage-lock" is only active on lowest multiplyer, correct ?
On all other multiplyers is lock-free ?


Perhaps there is still oportunity to increase battery life after all. I mean, if you use laptop on battery power and do something that require more power than the lowest, there is still some battery gain if you undervolt, correct ?

 

On the C2D all multipliers have a lower limit of 0.950V (on my T7200). I'm guessing it's the same with CD processors.

 

Correct. However, I myself wouldn't run power-expensive simulations on battery power unless I had to.

Right.

 

Hey E.B.E, excellent post

I did some undervolting myself to see what results I would get with my C2D T7200 (dell in sig), 2.0 ghz.

This processor goes from 6x to 12x multiplier, lowest VID at 6x is .950v (unchangable obviously)

Default voltage is .950v at 6x and 1.250v at 12x

I used I8kfangui to manually control the 2 fans in the computer instead of the BIOS dynamically controlling the fans according to temperature. I ran Prime 95 and SuperPi at the same time to get 100% CPU load on both cores.

1.250v (default)
Both fans at slow - stabilized at 68 C
Both fans at fast - stabilized at 63 C

1.150v (.1v under)
Both fans at slow - stabilized at 61 C
Both fans at fast - stabilized at 58 C

1.075v (.175 under)
Both fans at slow - stabilized at 56 C
Both fans at fast - stabilized at 55 C

I took manual control of my fans because otherwise, the BIOS will automatically try to maintain a certain temperature and toggle the fans between high/low/off, so I felt that comparisons at the same fan speeds would be the best "apples to apples."

I agree that undervolting significantly decreases operating temperature, especially at full blast. I should also point out that because the intermediate VID's are also lowered, it can increase battery life in a CD/C2D machine as well because the CPU is not necessarily chillin' at 6x all the time. (Unless you have your computer locked to only 6x on battery -- in which case you will see no difference).

Was 1.05 your lowest stable undervolt? I'm a bit reluctant to go under 1.075.. I'd rather not BSOD and very few C2D owners have posted their stable undervolts as a frame of reference.

 

10C temperature decrease is very similar to what I saw undervolting the P-M in my Z33. That was the very first thing I did out of the box

 

1.050 is stable here, at least as far as now. I didn't try to go lower, I don't have time for the trial-and-error procedure needed to determine the "lowest" stable voltage.

 

Dual Prime95 stable at 1.025V for many hours. I gave it a little room and am running at 1.037V (RMClock). I went down in steps with Prime95 until it started to give errors (reported by Prime95), come back up and 1.025V was the first stable point.

Someone awhile back was able to go all the way down to 0.950V at full speed with a Core Duo. It's half luck of course.

 

I am running a CoreDuo T2250 at full speed 24/7 (Folding@Home) along with CAD and Visual 2.0 daily, with zero problems. I have my voltage currently set @ 0.9875V. I have run two instances of Prime95 (to properly load both cores) with the Blend, Small FFT, and Large FFT for over 14 hours with zero trouble. I guess I lucked out and got a good processor. Keep in my mind I did all these test prior to tearing my notebook apart and replacing the thermal paste with the excellent Arctic Cooling MX-1 paste.

 

I tested my T5600 (in Q35) running 2 x Prime 95. One threw an error and stopped at 0.975V so I put the voltage back to 1.0V and did an overnight soak test with 2 x Prime 95. No problems then or thereafter.

I did a similar process for my T7200 (in X60 plus). I think it gave an error at just over 1V, so I had it set on 1.025V. Then, a month or so back I was doing some modelling which loaded both CPUs for hours and got a BSOD so I pushed the voltage up to 1.05V and have had no further problems.

The big advantage of using the stress test module in Prime 95 is that the built-in error checking will usually detect a problem before the BSOD stage. Prime 95 then stops. I have used two separate instances of Prime 95 to fully load both cores. However, I have read somewhere there is now a version which loads all cores.

John

 

Very good article indeed. I have done a very similar tests and graphs with pentium M and the results were the same. Temperature drop in game was around 13 degrees C. That is even higher than CPU at 100% load only because usually the heatsink covers both CPU and GPU (x700 in that test). GPU at 100% easily hits 85 so it actually heats CPU more than it would be at this load. With undervolting the whole story is much much better since the heatsink can do its job with ease in these situations and the cumulative heat can be easily brought out.

Cheers,

Ivan

 

We really ought to sticky an undervolting thread; up until now, I never though that could increase battery life besides underclocking the cpu, reducing screen brightness, etc.

As of late I've been keeping my T7200 (f3jp asus) at -0.1 baseline, and at 1.062 at 12x multiplier. If it's still good while I do my work in Illustrator, I'll probably lower it. It's been stable so far while gaming, and the battery life has gone up rather nicely. I think I can crank out 3 hours easily, instead of the 2.5 hours I'd usually get while surfing the web.

Also, the temperature is at a stable 37c instead of 45+; that's pretty awesome, considering that I have no thermal paste or desktop fans. It makes me more assured in that this machine won't die out in a few years

 

Can you undervolt the Core 2 Duo at lowest multiplier? (-0.1 baseline) You can't do that for the Core Duo.

 

Woops! That's a negative; 1.000, not 0.1! That would be instant suicide...

 

i hope i'm in the right spot to ask this question. i am interested in tweaking my dell e1405 notebook. I just recently purchased this, and am used to my desktop and its flexability with hardware tweaking. can a laptop be overcloked? Any suggestions? or should i go somewhere else?

 

Laptops are usually underclocked. It is recommended that they should be underclocked to save power and reduce heat/fan effort when the CPU is not used at full blast. Underclocking is by default. By underfclocking you shouldn't understand the CPU runs always below its rating. No; the CPU adjusts its speed dynamically to satisfy the needs of the task at hand.

About overclocking to increase the performance of the CPU at full blast, I don't know. But I certainly don't recommend it.

 

Anyone know why intel locked the lowest voltage multiplier for Core duos ? Undervolting cannot damage the hardware right ?

 

I don't know... I imagine some obscure marketing reason like being able to sell the low-voltage chips... Or there might be some engineering reason behind it.

I don't think it can damage the hardware, not when it's done one step at a time. Even if you're going too low, the worst that can happen is that the computer locks up, you restart in safe mode, and change the settings of or remove the underclocking utility.

 

IF there is the option of setting the voltage in the BIOS (like i've seen on desktop PCs), is this a better solution than using a program?

 

AFAIK there is no such option. It would be better at least because it would reduce overhead on the system.

 

I can't underclock my T7300 on my dell D630.

anybody tried that?

NHC simply doesn't allow me to change ANY settings...

 

Same for my T5200 so I use RMClock, very good.

 

Yes, I also suggest RMClock, NHC has some compatibility issues either with Vista, with newer chipsets, or both.

 

I've got a T7300 and I'm a bit confused. I've been playing around with RMclock and NHC and I don't really understand how they manage to accomplish different things. With NHC I can't manually undervolt at all for some reason. However, if I choose battery saving mode, i get an underclock to 800mhz and undervolt to 0,900 - which basically results in the CPU fan being completly silent (it always is as long as CPU stays under 50C). However, this mode is not dynamic, so I have to manually switch this setting if i want a higher cpu clock.

In RMclock however, I can choose the voltages myself, but here instead I can't reach 800 mhz, but instead 1200 mhz as lowest (dynamic though) and the lowest I can go here is 1,000 volts. Meaning it gets hotter in idle and the fan is on constant (not loud ofc, but im trying to make it optimal here ). RMclock also allows me to undervolt to 1,075 while loading the CPU maximal.

So I'm wondering if its somehow possible to enable dynamic mode, make the lowest 800mhz at 0,900 volt, and the highest 2000mhz with 1,075 volt, with 1 program? Thanks

EDIT: I just realised I can even change the clock directly in NHC taking it down as low as 90mhz if I wanted :O But for some weird reason the voltage controls in NHC are darkened out not letting me do anything..

 

they are darkened out for compatibility reasons, I guess.

I believe NHC uses a feature called CPU throttling, which is able to clock down your CPU past what you can achieve with the lowest multiplier setting. Google for more information. Throttling can be used in RMClock as well, it needs to be explicitly enabled (on the profile pages, "Use throttling (ODCM)" checkboxes).

I have tried it but I found the computer not so responsive when throttling is enabled, so I disabled it.

I don't know how the voltage settings are changed with the throttle setting, in RMClock, you're on your own there.

 

My T7300 also has 1.0V as the minimum selectable (but others may be different). The 800MHz mode is actually achieved by halving the FSB speed and then running the CPU at 8X but with a lower voltage (maybe 0.1V below the normal minimum?). It seems that at present RMClock may not support this mode and it may not happen until the author gets a computer with the new hardware. What do you see in CPU-Z when the computer is on idle?

Here's an example:


John

 

I did nothing to undervolt mine, but after one hour or so of very light usage (Abiword only opened) Cpu-z says that my T7300 is running at 800 MHz (or was it 600 MHz - 100 x6?) with a voltage of -lo and behold - 0.85 V.

Core-temp don't seem to work properly, since it fails to recognize the cpu as a 7300 and gives me a much lower clock speed (soething in the range of 200-300 Mhz) with a much higher voltage (1.295, IIRC).
I do not have snapshots since at the time I had uninstalled the TabletPC accessories (I did not know the snapshot tool was in there ).

Who should I believe?
Coretemp reports a temperature of 37-39°C for the two cores.

[/QUOTE]

 

OK. That's another T7300 going down to 0.85V. I can't check my current T7300 until Zepto sort out the BIOS.

I would use RMclock to monitor the temperatures. It can do this without changing the power management. If you do load RMclock, what voltage range does it offer? (The minimal and maximal voltages on the CPU info page). Mine is 1.000V and 1.250V.

You can capture the active window with Fn+Alt+PrtSc (press all three together) while Alt+PrtSc captures the whole display.

John

 

Ok, here are the screenshots. It seems I can go from 0.850 to 1.250.
Not sure at what frequency my cpu was running. It says 1600 GHz, and then it is throttled to 800 MHz. I take it this is what Cpu-z reports as 800 MHz operation.

 

It looks like you have 0.937V as the normal minimum, 1.25V or 1.275V as the maximum (your second shot shows 11 x @ 1.275V) and 0.85V for the 800MHz setting.

John

 

With no program running it will sometimes go down to 800mhz with 0,9 volts. However, whenever it is running at anything other than 800, it will always use 1,25 volts - which is most of the time. When I use RMclock it never goes to 800 even if i used performance on demand however. I've simply decided the best thing to do is run it at 2000mhz constant with 1,063 volts, that even seems to produce less heat and makes it more stable, than letting it throttle every halfsecond and switching volts all the time.