Is PowerTop Still Useful For Extending Your Battery Life?

[Ragas]

I can't find the link right now, but IIRC that's true only if you look at an isolated CPU. There's more components inside your computer.

If transistors cannot function your CPU will output random nonsense. Writing weird memory values causes your programs to crash. Violating bus protocols opens up an entirely different class of failures. I doubt that consumer grade hardware is meant to fail gracefully there.

Also, your CPU gets signals from everyone else, and those are at a fixed voltage. There you'll see an increased voltage differential when undervolting your CPU which may or may not harm whatever's responsible for the conversion.


So yeah, slightly undervolting will probably result in system crashes before it physically damages anything. But if you lower it too far and got a bit of bad luck, you may end up with broken hardware.

I know what you mean. there were some vague reports of some Overclocker Ram destroying Undervolted AMD CPUs because the the converter wasn't constructed for such for those voltage differences.
But it also was never confirmed and thats a very rare cause where the RAM was running on too high voltage and the CPU on too low Voltage.

Actually this would be the fault of the overvolted RAM and not the undervolted CPU because the CPU gets constructed for specific voltgaes too hold and that must always be the maximum specified volages. because on peak when switching you can always have a minimum time when it goes up from minimum to full in an instant.
So in that case the hardware was already running out of specification and the undervoltig just gave it the rest (eveb though it's not is's fault).

So you are right in a way, that if you are using Undervolting you could rip open already existing hardware problems and thus sophistically kill your hardware.

But that won't ever happen to you because: Todays CPUs already use undervolting. they won't permit voltage changing under specified levels without really deep hacking. Using PHC on CPUs which don't meet those prerequisites makes no sense since they have no dynamic voltage scaling which is nonsense from every single aspect.

And the thing that at too low voltages the CPU might put out wrong results true is your own risk. But it won't kill your hardware.

[xianthax]

I can't find the link right now, but IIRC that's true only if you look at an isolated CPU. There's more components inside your computer.

If transistors cannot function your CPU will output random nonsense. Writing weird memory values causes your programs to crash. Violating bus protocols opens up an entirely different class of failures. I doubt that consumer grade hardware is meant to fail gracefully there.

Also, your CPU gets signals from everyone else, and those are at a fixed voltage. There you'll see an increased voltage differential when undervolting your CPU which may or may not harm whatever's responsible for the conversion.


So yeah, slightly undervolting will probably result in system crashes before it physically damages anything. But if you lower it too far and got a bit of bad luck, you may end up with broken hardware.

CPU's don't operate at 1 voltage anymore, hell not even all embedded SoC's do. The the I/O portions have a supply at one voltage and the core operates at another and in some cases there are various voltages for various sections of the die. The voltages never really interact, the various portions of the chip are buffered from each other.

As for the P=IV crowd:

The actual danger is higher voltage isn't a result of this equation but rather that increasing the voltage increases leakage current in modern transistors which is responsible for most of the heat generation(wasted energy) in modern semiconductors, its been many years since leakage current wasn't the major problem in asics built on small process sizes.

[arjan_intel]

I wonder where Michael looked for PowerTOP git...
probably in the wrong place ;-)

while Auke has been on baby leave for 2 months, we've done quite a few things before then and for example MeeGo ships what effectively is version 1.12.

The biggest new thing is finding those applications that keep your disk awake.

Also... what Michael did not write about in this article, is if PowerTOP identified applications that woke the system up a lot (more than once per second)...

While maybe end users can't do much with that information (other than potentially uninstalling or not starting said applications), the developers of applications and distributions can do a lot with this. For example, in MeeGo, PowerTOP is our number one tool for diagnosing any power issues, and we run it regularly to find and fix things that wake the system (cpu and/or disk) up.

I realize that this is not sexy end-user usage... but for developers it's key.
I'm glad that for normal users there's a lot less to do, that means that linux now does many more things right by default in various distributions....

[yotambien]

I don't usually high-five people for doing their job or devoting time to their hobbies. But in this case I want to say that what you guys did at Intel with PowerTOP and Linux power management in general is absolutely terrific. If memory serves, between more or less 2.6.17 and 2.6.21-2 there was a revolution in terms of power savings, and not only from the kernel. PowerTOP put to shame the modules and applications that were misbehaving, forcing developers to fix them. Suddenly everybody was checking where those extra 60 wakeups came from, suggesting workarounds and testing patches to improve the situation. Getting your system to iddle at less than 10 wups/s was quite a challenge and a fun thing to do.

I don't know whether it's too much to say that there was a change in awareness, attitudes and coding practises as a result--I couldn't possibly know that much. But that nowadays PowerTOP doesn't spit anything terribly nasty in a default installation and power@bughost is a very low traffic mailing list is a sure sign of success.

[Xake]

I wonder where Michael looked for PowerTOP git...
probably in the wrong place ;-)

while Auke has been on baby leave for 2 months, we've done quite a few things before then and for example MeeGo ships what effectively is version 1.12.

Well, according to git.moblin.org the only changes this year is "l10n" (mostly spanish). Also is there any plan to roll out a 1.12 tag/tarball?

[sofar]

Well, according to git.moblin.org the only changes this year is "l10n" (mostly spanish). Also is there any plan to roll out a 1.12 tag/tarball?

[ahkok@localhost powertop]$ git diff v1.11.. | diffstat -p1 | grep -v po\/
.gitignore | 2
Makefile | 16 -
ahci-alpm.c | 266 +++++++++++++++++
alsa-power.c | 202 +++++++++++++
bluetooth.c | 6
config.c | 5
cpufreq.c | 2
cpufreqstats.c | 17 +
display.c | 17 -
dmesg.c | 63 ++++
misctips.c | 91 ++---
perf.c | 273 +++++++++++++++++
perf_event.h | 863 ++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++
powertop.1 | 38 --
powertop.8 | 38 ++
powertop.c | 212 ++++++++++---
powertop.h | 28 +
process.c | 4
sata.c | 7
suggestions.c | 6
urbnum.c | 12
usb.c | 39 +-
wifi-new.c | 205 +++++++++++++
41 files changed, 6838 insertions(+), 2562 deletions(-)


yeah, it's time I do a new release

/auke

[molecule-eye]

The only thing I really use powertop for is to determine which things are causing the most wakeups. Once determined I quit or disable these things when I run on battery power. You also learn quite a few interesting things about your software. E.g., I notice that Amarok causes a huge amount of wakeups when paused but not when stopped--go figure! So on battery I now always make sure to stop songs rather than pause them. I've also noticed that disabling the selection scanner of mdic causes fewer wakeups, so I do this too just before going on battery. In the end, I'm able to squeeze out as much battery life on my Kubuntu install as I can on my Windows install.

[Craig73]

I notice that Amarok causes a huge amount of wakeups when paused but not when stopped--go figure!

Did you log a bug for this?

[molecule-eye]

Did you log a bug for this?

I didn't, but someone else had: https://bugs.kde.org/show_bug.cgi?id=235165