Skip to main content

Slow and Steady: Undervolting your CPU


The following only applies to Windows machines. There may be something similar for Mac and Linux, but I didn't look into it.

Why: It makes the CPU last longer, extends battery life, and makes the laptop quieter (the fan runs a lot less).

Risks: None - just the time it takes to do it.

Why it works: A standard voltage range is applied to all chips of a given type that a manufacturer makes. Each chip may have a different tolerance where it is stable, but by setting the voltage high, all of the chips in a given run will remain stable. Your chip may only be able to handle the max voltage, or it may be like mine - the min voltage for all clock speeds.

I followed this guide's advice for a while (drop the voltage by 0.025 & test for 45 minutes), then read down a ways the following comment:

I just keep the CPU stress test running and drop the voltage at the maximum multiplier at about 1 step every 15 seconds starting at 1.10V (for Intel Core / Core Duo). This very quickly finds the voltage which is definitely unstable and BSODs (keep a recod of the voltage as you drop it). I then put the voltage back up two steps and run an overnight stress test.


I ended up getting down to the minimum voltage in just a few minutes & ran a test for 45 minutes. It was stable, so I left it there. I'll run an overnight stress test this weekend.

The basics are to run three utilities together:
1) CPU Monitor to tell me the temperature of the cores.
2) A utility to make the CPU max out.
3) A utility to change the CPU voltage for each clock speed.

Keep dropping the voltage until you see the Blue Screen of Death, then kick it up two notches. You change only the max clock speed voltage as you go, then change the others once you have a setting for that (my chip has 6 steps). Once you're happy with it, you set the utility to kick off at startup (it runs once & goes away). You can always boot to safe mode & get factory defaults this way - safer than doing it at the BIOS level.

(Thanks to Uwe Hermann for the photo of a Celeron CPU)
Post a Comment

Popular posts from this blog

Agile Performance Management: Why Performance Reviews Suck

Many thanks to Mary Poppendieck, who wrote about this topic in 2004, and proposed a comprehensive solution.  She is the inspiration for much of my thinking on this subject.  She is also a better writer than I am a cartoonist.


Performance reviews suck.  I don't know of anyone who goes into their appraisal without some trepidation.  Your boss is guaranteed spring some surprise criticism on you that is ill-informed or misses the point as you see it.  It's a real challenge not to get defensive about that.

The only thing that makes your own performance review suck less is having to give them.  As a manager, I have dished out quite a few, and some of them went pretty badly.  (To the people at my first management job: Thanks for helping me learn how to get better at them.  Your sacrifice was not in vain.)  Since then, receiving one isn't nearly as gut-wrenching, if only because I try to make it easier for the guy on the other side of the desk.  I've been there, and I know how …

Do. Not. Optimize.

You've probably heard this quote before:
Premature optimization is the root of all evil.
 - Tony Hoare
Speculative optimization is always wasted time.  In the absence of an actual performance problem, you're just burning time that could be better spent on refactoring your code to make it clearer.  This is exacerbated because performance-optimized code is usually harder to read than code which hasn't received such treatment.

Here is what you're doing when you optimize:
Adding code that now must be maintained.Obfuscating the existing code.Spending time writing code that doesn't add value. But what's that you say?  You have the experience and know-how to decide when optimization is needed?  Maybe, but probably not.   The people at Sun and Oracle may or may not be smarter than  you or me, but they certainly know more about optimizing Java bytecode than we do.

For example, some people think that having a large number of classes is slower than the alternative.  This …

Showing Off: How to Do a User Demo

Rather than repeating what has been said elsewhere, here is a nice short post on agile-for-all that covers the basics.

Here are a few things for my own future reference and teams that I'm working with...

Try to keep each demo to 5 minutes or less.   If it's longer than that, it's possible that you should be demoing more than one story.  More likely, you're just being too wordy.

TALK LOUDLY.   No, louder than that.  Louder.  Do you feel like you're yelling?  OK, that's about right.  You need to put your voice in public-address mode for 5 minutes.

Focus on why your audience should care about the story  This is particularly important for back-end work.  For example: Your story generates a feed of XML that will be consumed by another application. Show the output, and point to a couple of salient features in it.  Then be done.

The important part of the above is "show the output."  Showing the end users how to interact with your service is a separate sit-d…