Undervolting as a means of CPU Cooling

burcakb · Post by **burcakb** » Tue Feb 08, 2005 10:58 am

Although there are several references to undervolting throughout the site & forums, I see few newcomers taking advantage of this wonderful & free cooling technique. Then I noticed that most articles talk about undervolting & underclocking in one breath and that may be a turn-off for newcomers.

Anyway, while I was preaching the wonders of undervolting in another forum, I received the usual baseless challange that undervolting would make my CPU unstable/slow/<insert favorite accusation>. One slightly open-minded zealot ventured to try it if I would run SuperPi for 32 Million digits at low & default voltages and compare - only then he'd be convinced. I did it and the results are impressive enough to post and point newcomers to. As "seeing is believing", I thought I'd share it with non-believers

Here it is.

First a warning: I did this on the spur of the moment and I realise there are a few parameters that could affect the outcomes. However, the variations would be small and in no way account for the astonishing temperature gradient fully.

Test Bed:
The Thor system specified in my sig.

Test method: Start Motherboard Monitor, run SuperPi (32M), note time taken for calculation & high-low table from MBM, change voltage, reset MBM statistics, run SuperPi (32M) again, note time taken & high-low table from MBM.

Test CPU: Athlon64 3000+ Newcastle core (130nm), running at 210x10 = 2104 MHz clock speed.

Room ambient throughout test: 24C

First test is with the CPU core voltage at 1.2V:

Time taken by SuperPi : 39 min 33 sec.
MBM Data:

Code: Select all

+-----------------------------+----------+----------+----------+----------+
|Sensor                       | Current  | Low      | High     | Average  |
+-----------------------------+----------+----------+----------+----------+
|Case                         | 36° C    | 36° C    | 37° C    | 36° C    |
|CPU                          | 46° C    | 46° C    | 48° C    | 46° C    |
|PWM                          | 45° C    | 45° C    | 47° C    | 45° C    |
|CPU Core                     | 1.20 V   | 1.20 V   | 1.21 V   | 1.20 V   |
|CPU Fan                      | 120 RPM  | 0 RPM    | 421 RPM  | 146 RPM  |
|Exhaust Fan                  | 722 RPM  | 662 RPM  | 783 RPM  | 753 RPM  |
+-----------------------------+----------+----------+----------+----------+

Second test is with the CPU core voltage at 1.5V:

Time taken by SuperPi : 39 min 32 sec.
MBM Data:

Code: Select all

+-----------------------------+----------+----------+----------+----------+
|Sensor                       | Current  | Low      | High     | Average  |
+-----------------------------+----------+----------+----------+----------+
|Case                         | 37° C    | 36° C    | 37° C    | 36° C    |
|CPU                          | 78° C    | 59° C    | 78° C    | 75° C    |
|PWM                          | 66° C    | 50° C    | 67° C    | 63° C    |
|CPU Core                     | 1.48 V   | 1.47 V   | 1.51 V   | 1.49 V   |
|CPU Fan                      | 180 RPM  | 0 RPM    | 361 RPM  | 160 RPM  |
|Exhaust Fan                  | 783 RPM  | 662 RPM  | 783 RPM  | 763 RPM  |
+-----------------------------+----------+----------+----------+----------+

As can be seen, undervolting has no effect on either the stability or the speed of the CPU. However, the heat differential is enourmous. 32C difference just on voltage change.

I should note that my CPU fan is slowed down extremely (ignore the fan rpm, uGuru cannot record reliably below 600 rpm). It is one factor for the large temperature differential, but it also allows quiet computing and this site is all about quiet

Cooling in between the tests would make a slight difference in temps.

Heat produced is proportional to the SQUARE of the voltage applied.
So: Vlow^2/VHigh^2 gives us the average reduction in heat. In the above test, it's 36% reduction

Wraith · Post by **Wraith** » Fri Feb 11, 2005 7:37 pm

I hadn't tried any undervolting at all until the last couple weeks. I had been running my 1700+ at 1.5v (default) and 182 FSB (~500 MHz overclock), and load temps were around 54C. I set my FSB back to 133 (default), and dropped to 1.35v. Load temps around 40C.

Tonight I dropped down to 1.15v, and am getting around 36C load, running Prime95. Completely stable for 2hr 26min of Prime95. Fan speeds were the same throughout (back 5v Papst, front Nexus 5.5v, CPU Panaflo on fanmate, PSU fan dialled to minimum). I'm definitely pleased with the results so far.

Cams · Post by **Cams** » Sat Feb 12, 2005 1:26 am

The proof most certainly is in the pudding. It adds another tool to our cooling toolbox.

Just as a matter of interest, what can one do if their motherboard does not support undervolting through the BIOS. I tried CrystalCPUID on my Gigabyte GA-8ITX and there didn't seem to anything there that would allow me to undervolt. Am I stuck with the voltage that I'm given?

Denorios · Post by **Denorios** » Sat Feb 12, 2005 6:04 am

I've tried overclocking/volting and underclocking/volting on various occasions, and I agree that the results can be quite dramatic. Undervolting/clocking can seriously reduce heat without compromising performance. The major concern is stability, which is difficult to measure, and depends upon what level of stability you are willing to accept. Some will accept reasonable, other demand rock-solid.

My concern in this case is that you only ran SuperPi for 40 minutes. When testing, I tend to run Prime95 at maximum stress level from the time I go to bed to the time I get back from work the following day - at least 18 hours. I have seen errors crop up after 11 hours running. I would strongly recommend much more extensive testing before you settle with these settings.

By the way, what is the default voltage of the 3000+?

ckolivas · Post by **ckolivas** » Sat Feb 12, 2005 6:13 am

I'm still looking for firm vidpinning results on P4s to undervolt my motherboard. The ASUS P4PE simply doesn't support undervolting as mentioned in topics from way back. Doesn't seem like there are confirmed reports of successful safe vidpinning for undervolting as far as I can tell.

burcakb · Post by **burcakb** » Sat Feb 12, 2005 7:22 am

Denorios,

I was specifically asked to use SuperPi and 32M digits is the max it can do and my CPU crunches it in 40 mins. I went on ahead to prove that Prime95 and CPUBurn stressed the CPU more and that Prime95 was a better test, etc. I always Prime95 for stability.

Athlon64 3000+ Newcastle core default speed is 200 x 10 = 2000 MHz.

btw, anything other than rock-solid is unstable IMO. Otherwise, I'd just settle for 230x10=2300 which my CPU can do but my Radeon in 3D mode doesn't like.

Cams, you need to use AMD K7/K8 Multiplier management under Function menu OR you set up CnQ levels in Multiplier Management Setting under the File menu and then turn on Multiplier Management under Function menu.

Post by **Ralf Hutter** » Sat Feb 12, 2005 8:28 am

Denorios wrote:I've tried underclocking/volting on various occasions, and I agree that the results can be quite dramatic. ........The major concern is stability, which is difficult to measure,

No it's not. Prime95 is the stability-checking app of choice for overclockers, and it works great for checking undervolted systems as well. Undervolting is basically overclocking in reverse.

Denorios · Post by **Denorios** » Sat Feb 12, 2005 8:47 am

Yes, I know. The point I was making is that Prime 95 can run for hours before failing, so a short test doesn't mean a lot. I wouldn't really consider a system stable until it had passed at least 12 hours. Equally, Prime 95 has been known to crash for reasons that have nothing to do with the CPU. And some people are prepared to accept less than rock solid.

Edit: I was just slightly concerned because 40 minutes is not a lot of testing for such a large voltage difference.

Cams · Post by **Cams** » Sun Feb 13, 2005 12:56 am

burcakb wrote:Cams, you need to use AMD K7/K8 Multiplier management under Function menu OR you set up CnQ levels in Multiplier Management Setting under the File menu and then turn on Multiplier Management under Function menu.

All of those are greyed out

Wedge · Post by **Wedge** » Sun Feb 13, 2005 12:17 pm

Hmm.....reading this thread makes me want to try it. I have a 2.8C at 3GHz; the voltage is 1.6, if memory serves me correctly.

Regardless, thanks for the info burcakb. Very interesting (and quite dramatic -- I had no idea).

The Instigator · Post by **The Instigator** » Sun Feb 13, 2005 2:43 pm

It defenitely works. Im currently running my 2.4C @ 3.0 with the voltage at 1.525 which is the lowest my MoBo will go. I just finished re-encoding about 4 DVDs to DivX without a hitch. When I run at 3.3Ghz however, It will reboot occasionally without warning even with the voltage increased to 1.7V. Who would think 25Mhz FSB settings could make such a difference in stability and temps?

On the other hand, I have tried lowering the FSB to 100mhz (1.2Ghz on my processor) and lowest voltage settings and my temps didnt get any lower than they do with the CPU at normal or overclocked speeds. My guess is that the XP-120 is doing its job VERY well.

Wedge · Post by **Wedge** » Sun Feb 13, 2005 5:04 pm

The Instigator wrote:My guess is that the XP-120 is doing its job VERY well.

Instigator, I see we have the same motherboard too. That's good to know if I buy an XP-120 in the future.

What voltage does your mb show with cpuz?

BigDu21 · Post by **BigDu21** » Tue Feb 15, 2005 10:57 pm

hey burcakb good idea for a thread.

*edited//new question**
I tried to undervolt my cpu, dropping it from 1.7 (which actually shows in 8rdavcore as 1.61) to 1.675, then running prime 95. When i run prime 95 it gives me an error after 5 minutes complaining about rounding--has anybody seen this before? is it a hardware problem or a setting that i can fix?

Specs:
Gigabyte 7N400 Pro, AthlonXP 3000+, 2.09 Ghz, 400 Mhz FSB. Thermalright SI-97 cpu cooler w/ panaflo 92mm. 1 GB 400Mhz RAM, GF6800 GT w/ artic silencer. Also 2 WD HDDs, a Plexor712A, and a SB Audigy2.

Badger · Post by **Badger** » Sun Feb 20, 2005 10:13 am

The error means you're not giving your processor enough voltage. Up the voltage back up to stock, run Prime95 again, see if you get the error, if you do, up your voltage another notch and continue this pattern until Prime is stable, the processor gets too hot, or you find something better to do.