Due to the fact that the original Link by Borkar [1] is down and the link to the archive by Daytona [2] is slow to respond I decided to copy the contents here. The original author is Ikovac of thegamersbook.com. He deserves all the credits.
[1]
http://www.thegamebooks.com/undervolting-intel-pentium-m---how-to-keep-your-gaming-cool-t78.html
[2]
theGameBooks.com • View topic - Undervolting Intel Pentium M - how to keep your gaming cool
Ikovac wrote:
Post subject: Undervolting Intel Pentium M - how to keep your gaming cool
PostPosted: Tue Jan 10, 2006 11:57 pm
Many people ask about voltages and settings for a particular processor.
The general thing is that a higher CPU voltage results in a higher stability at overclocked speeds. But it is also true that default voltages are way higher than needed in most systems.
Now don't get me wrong, I don't say that ALL CPUs can and will work at higher or lower voltage. It's not very important but if you do want to know more about this fact, then I'd advise you to read this. There is a different tool used and the procedure is more "scientific".
I think it is a good reading, but probably not the only way of doing it.
Try google, undervolting is a quite common thing and as far as I know you can undervolt AMD processors too.
In this article I will focus on Notebook Hardware Control as the Intel Pentium M processor undervolting tool and normal work and gaming as a stability test. The goal is not to get the lowest voltages, but to show the general idea and prove it offers positive results.
So let's go!
Stability
Stability means no errors. Errors can be the result of a CPU beeing too fast, the voltage beeing to low or to high or the temperature.
In silicon chips higher temperature means lower resistance. Lower resistance means higher voltage and thus higher current. That results in higher output power in the form of heat. So it heats faster and faster, temperature raises exponentially faster thus endangering the processor itself very fast. Manufacturers set the best voltages that are stable in many different configurations and situations.
So, is the voltage and current very precisely adjusted to fit the needs? Didn't Intel make the Pentium M CPU for my particular notebook?
NO! They gave their specs to the notebook manufacturers. I expect that my computer is well (I hope) in the tolerance margins.
But what if you could undervolt the CPU, and still get the same stability/performance with less heat?
Heat
Heat production is exponentially dependant on voltage and current. So if you lower voltage a bit, you get a big difference in heat drop. And vice versa. Note that chips cannot be overvolted or undervolted too much. There is an optimum range when logic circles work at best. So temperatures are also in some way limited by maximum and minimum voltages and heatsink ability to cool it. Heat is the main reason for aging of electronic devices. The lower - the better life expectance. Also notebooks are supposed to be nice and cool, or warm at most. Heat also produces the need for extra cooling-fans. Noise and batterylife are also a very important factor in notebooks, and the fan is raising the noise and reducing the batterylife - so you can do the math yourself.
Speedstep
Intel baby! Imagine a processor that could work at different voltages and multipliers of FSB. That is Speedstep. In less than 1/2000 of a second it can change speed (frequency and voltage) depending on the workload. 99% of time it works either on minimum multiplier/voltage or maximum. For PM750 it is 6x/0,988V and 14x/1,260V. All multipliers are already stored in NHC in the form of a table.
Tool
Notebook Hardware Control is probably one of the best freeware that ever existed for notebooks (especially with an ATi card). One of its great features is setting voltages for Pentium M processors. Celeron M unfortunately is not supported (at least the last notebook that I have seen couldn't set it). P4M is also a different story. Check CPU voltages. Can you access it? If you can - great! Read on.
slap-o: "If you can't access it that's no problem, there are lots of other tools you could use - but I'll write another article about that. NHC is great if you can use it!"
Rules
After reading tons of posts and articles I learned this:
1. You should be able to lower the voltage around 0,2 V per multiplier.
2. For PM7x0 series the stable minimum is: ~0,750V, the maximum is around 1,100V
3. You should check the desired lower voltage and then raise the voltage around 10-20% in order to greatly enhance stability.
4. Run extended tests for stability. Prime95 for example, but be sure to include DVD playback and record, plugging and unplugging (USB)devices and play some games. (I will explain this later on).
5. Be careful with OVERCLOCKING and UNDERVOLTING at the same time. Undervolting lowers stability, and overclocking too. Add 2 and 2. :eureka:
My procedure
I have set new voltages to my PM750 (1,86 GHz). Now it is interesting because I did it on an OVERCLOCKED machine.
My Uniwill 259EA barebone has a BIOS setting that overclocks FSB by 5%. It ends up in memory clock at 140 MHz and 1,96 GHz CPU.
Now that brings some noticeable speed gain. Especially in memory benchmarks. I can also use either built in GMA900 or X700 128MB card via switch.
I followed my intuition and experience gathered from others. I lowered max and min voltage by 0,2 V. Checked if it works (there is simple and advanced checking method already built in NHC), and then raised to some higher level. I didn't bother to set all multipliers. The time that CPU actually uses them is so small that I just didn't bother.
arp: I would do it eventually, but it would be just fine tuning.
I found 0,700 V to be stable, but that is not enough. I raised by 0,05 and now is 0,748V. No problems so far. 1,000V for max multiplier instantly resets my notebook. 1,100 V runs very fine. Now that is a bit higher than expected (many people have no problems at 1,052V). I just want to be extra sure while playing. Few days of working - no problems. So far so good.
Results
Idle temp on an Ati x700 with powerplay is the same as at default voltage. It balances between 51 and 56 when fan goes on and cools it to 51. Time between fan-activity is longer and temperature drops more quickly. I think the time between fan-activity is at least 50% longer (I didn't measure). Actually, the Ati GPU is responsible for this temperature behaviour, not the CPU. They are both under same heatsink and the GPU actually heats the CPU! Interesting. :hothead:
Idle on GMA900 is another story. Here GMA900 has very small or none at all impact. And guess what - idle temp is 52 (after few hours of surfing and working in windows). Before it would reach 56. Fan of course never goes on.
100% load on Ati x700 - now we are talking: top 65-67 degrees celsius in comparison to top 75-77 before. My fan shifts gears at every 5 degrees. This undervolting is keepeing my fan TWO speeds down. I think thats very good! That's where you can see the true power of undervolting.
100% on GMA900 reaches 56 - fan turns on and that is it. First speed, never reaches 60. Before it could easily reach 66.
To illustrate this here is a diagram showing undervolting temp differences at 1,96GHz (OVERCLOCKED). If I overclock ATI x700 GPU and MEM and set latencies (another story) i get very similar results for ATI configuration.
Image [sorry, not available anymore]
Battery time was prolonged on Ati - 10 mins. On Gma900 - 20 mins. That is OK with me considering the "not-so-good battery" I have. :-/
Dealing with problems
1. Undervolting shouldn't be dangerous. Keep it in 0,1 - 0,2V limit and you should be pretty safe. More is welcome if you know what are you doing. Overvolting is much more dangerous.
2. Instability when plugging in usb devices (or some other), playing DVD, etc... Well this is experienced with very low voltages (that is why is recommended to set 10-20% higher voltage once you find the stable low value). I guess that happens because the CPU is just too prone to errors when extremly undervolted. In NHC documentation is a paragraph about voltage fluctuation. Not only that the mainboard cannot actually set exact voltages, there is another problem.
3. Instability when on battery. I think it can be connected to 2. I have a experience from one forum member that on battery the CPU becomes more prone to errors. A small raise in voltage solves the problem.
Imagine your Mainboard, CPU, USB hubs and drives as junction devices on a net made of pipes. Each pipe represents a conductor and pressure in it is voltage (that you just have lowered). There are like thousands of them. USB hub is on one end, power cord or battery on the other. Main power keeps (pumps) pressure to all junctions and devices and keeps voltage constant. The flow of electrons goes through devices into ground making some work along and dissipating heat. Now you suddenly connect a new external USB hard disk. For a brief moment it behaves like a short circuit making a sudden and quite strong pressure-voltage drop in one part of the net. Although USB hub has its own way of dealing with it (capacitors), and many other regulatory devices come in dealing with this drop, it is almost certain that voltage regulating circuits that are responsible for CPU will feel the effect. As a result CPU voltage will drop a bit for a very short time. Maybe enough to restart your comp or show BSOD. That is why you need this 10-20% safety level. The same is with DVD - keeping constant stream of data, and synchronizing head with different tracks on the disc produce same disturbancies in the voltages. On the other side, power source is stabilized, but still produces small differences in voltages too. Some configurations suffer a great deal, some not at all. In most cases, 0,1V or 10-20% higher voltage solves all stated above.
Conclusion
Undervolting on notebooks (and silent desktop PC-s) is a good thing. I would add it is very useful too. No big knowledge is needed and a little voltage change produces a quite big temperature drop.
And believe me, in a notebook that's always a good thing!
Better battery times (people report up to 30 min or more) and an even more silent device!
Any risks? No! I have never heard that somebody killed his or her processor by undervolting. I guess it can be done but only if you are extremly unlucky, or you simply don't know what are you doing. In any case you do this on your own risk. If you undervolt too much the worst thing that would be is a reset and that can pull other things along. Just be careful, that is all. But if you need any help, we're here to help!
Happy undervolting,
Ivan
Problems? See this thread at archive.org.
