Ubuntu's Power Consumption Tested

RedDragon writes "Ubuntu 7.10 is due out on Thursday, October 18, and in addition to desktop 3D effects, GNOME 2.20, and other features is the use of the Linux 2.6.22 kernel with the tick-less (CONFIG_NO_HZ) kernel feature. But does this mean enhanced power savings when compared to past Ubuntu releases? Phoronix tested Ubuntu power consumption looking back 2-1/2 years at the six Ubuntu releases from Ubuntu 5.04 to the yet-to-be-released Ubuntu 7.10. Testing was done when the system was idling and then under load, and when the Lenovo notebook was powered via the battery and then again with the AC adapter. The Pentium M CPU temperature was also monitored. While Ubuntu 7.10 does include the tick-less kernel feature, more daemons and processes running by default on these modern Ubuntu releases is actually causing an increase in power consumption."

Is this supposed to be a surprise?

[cleatsupkeep (1132585)]

I'm confused - don't computers now use more power then they used to? Because of new software and being able to do more powerful things?

I mean - Vista will use more power than Windows XP, OS X will use more power than Mac OS 9.

Or is there a fundamental flaw in my logic that I'm missing here?

Re:Is this supposed to be a surprise?

[imbaczek (690596)]

Or is there a fundamental flaw in my logic that I'm missing here?

Yes, you are. Some parts are manufactured with power consumption being #1 priority and software is also getting smarter. (As TFA admits, at least theoretically :grin:>)

Re:Is this supposed to be a surprise?

[reset_button (903303)]

Some parts are manufactured with power consumption being #1 priority

Since all of the tests were run on the same hardware, power-efficient hardware is taken out of the equation.

Re:Is this supposed to be a surprise?

[Cruicky (1122359)]

It would have been nice if the article confirmed that the HPET timer was active, which I believe is rather important for the tickless kernel to work most efficiently.

Re:Is this supposed to be a surprise?

[dpilot (134227)]

The HPET stuff is now scheduled for merge into the 2.6.24 kernel. I've had to patch my earlier kernels to get HPET, which as you say is really necessary for tickless to do its stuff. The article suggests that this is a stock Gutsy installation. But then again, most distros do a bit of custom patching of their kernels. In particular, Gentoo does not include the HPET patch.

So the question here: Does the Gutsy kernel have the HPET patch applied?

If not, then these power numbers are definitely pessimistic, presuming that they move to an HPET kernel (2.6.24+) as it's available.

Someone here with a Gutsy system should run "powertop" on it, and let us know. IIRC, powertop suggested that I use the HPET, and with a little digging I found that a patch was needed, and took care of it.

Re:

[A beautiful mind (821714)]

HPET is in the vanilla linux kernel since at least 2.6.21, because I had it working after a motherboard flash update. The patches you talk about is actually helping to enable HPET support for some chipsets, but are not mandatory for a working HPET support.

Re:Is this supposed to be a surprise?

[SuperQ (431)]

Does this answer your question?

$ grep HPET /boot/config-2.6.22-14-generic
CONFIG_HPET=y
CONFIG_HPET_MMAP=y
# CONFIG_HPET_RTC_IRQ is not set
CONFIG_HPET_TIMER=y
CONFIG_HPET_EMULATE_RTC=y

$ dmesg | grep hpet
[ 8.328261] hpet0: at MMIO 0xfed00000, IRQs 2, 8, 0
[ 8.328266] hpet0: 3 64-bit timers, 14318180 Hz
[ 0.744000] Time: hpet clocksource has been installed.

Kind of.

[khasim (1285)]

Although each newer system does more ... they've also improved the code so that it does so more efficiently. Or in the case of the tick-less kernel, other code has changed.

So, the question is: Do the improvements offset the additional features.

The answer is: Yes, to a degree. 7.10beta runs cooler and more efficiently than 7.04 ... but still uses more power than even earlier releases did.

So the next question is: How many of the new features can you shut off because you do not need them and how much of a power savings will you see then?

All of them.

[SanityInAnarchy (655584)]

So the next question is: How many of the new features can you shut off because you do not need them and how much of a power savings will you see then?

Given that the stock Ubuntu (if you don't include "restricted drivers") comes with FULL source code, yes, all of them.

On a more realistic note, most people do need restricted drivers, and most people don't want to mess around with source code. But it's based on Debian, which means, for the most part, you can completely remove services you don't need, point and click, provided you know what they are.

Then again, I actually do want most of these services -- for example, the parts that make everything plug'n'play, from USB storage to wireless, even the CD "autorun" feature of Windows if you really want it. Most users won't have to think about "mounting" any more than they do on Windows, and somewhat more than they might on OS X, and that's a good thing.

Re:

[Daengbo (523424)]

Scary for me -- CDs auto unmount when you press the eject button. When did that happen? Worried I'll start getting BSODs next ...

Re:

[ozamosi (615254)]

Did you use Compiz in Gutsy but not in Feisty, then? A lot of people add Compiz to their Feisty installations, but it is default only from Gutsy, so I'm going to assume that's a "yes" in this post.

When activating Compiz on my laptop, I start to fear hearing problems, because the fans have to be at maximum speed non-stop (it's a Macbook, and I've been using it in my lap - any reproductive abilities are in other words long gone, so I don't have to fear that), while they are off at all times except when playin

Re:

[volsung (378)]

By going to smaller transistors, lower voltages, and more clever power management schemes, they have managed to get more work done per watt than before. A new 3 GHz Athlon64 X2 requires 89W of power, whereas the old 1.4 GHz Athlon Thunderbird used 74W.

Re:Is this supposed to be a surprise?

[gameforge (965493)]

Better efficiency would involve the second number being lower than the first.

No. Better efficiency would involve the second number being lower and also having equal clock speeds. The former is multiples faster as it has more than double the clock rate and is a dual core.

So if you start your number crunching computer program and push "Start" and it takes 15 minutes on the first CPU and over 40 on the second, presuming you were to turn your computer off when the program finished, you'd have used the first one for less than half the time.

With the same battery I'll be able to use my laptop for 20% less time (say 2.5 hours instead of 3). If it does more faster, how come I get 30 minutes less time to use it before my battery craps out?

Not really; most laptops and many desktops can scale their speed. If you want to accomplish 15 minutes of work in 40 minutes, you can either throttle the newer CPU's speed (presumably using much less than the full power rating) or replace it with an older processor which is not as efficient (and therefore equally as fast, but likely to use more power).

If you scaled your new one down to less than the speed of your old one, you'd get more time out of it. So if you're watching a DVD and not really accomplishing a lot of "work", that's what you'd do to get more time than the old laptop but still have more processing power.

Remember, 3GHz refers to CPU clock cycles per second - an old thunderbird gets less done in a cycle than a new Athlon64 X2. So even a 1.4GHz single core Athlon64 is faster than a 1.4GHz Thunderbird. So you can slow the new one down from 1.4GHz and still get the same work out of it. A DVD might be choppy at 500MHz on a really old machine, but a brand new state of the art processor might be able to deal with it just fine at 500MHz, even if both machines have similar bus and memory speeds and come with the same MPEG decoding video card.

What would be better is a CPU that can use up to 89W when it needs it, then falls back to much lower - say 10W - when it idle and waiting for me to type a clever response into Slashdot.

Actually, this is a characteristic of both transistors and vacuum tubes, and therefore literally all CPUs do this. The amount of voltage supplied to the CPU is supposed to be constant - but the more transistors you use, the more amperes are drawn (volts * amps = watts). Relative to peak power usage, the difference between two idle CPUs is likely negligible, even for older models.

Find a computer with a variable speed CPU fan, and listen for it to shut off when you're idling. Less heat means less power.

Re:

[volsung (378)]

You do realize that a dual core Athlon64 is many times faster than an Athlon Thunderbird? You can easily underclock an Athlon 64 until it uses less than 74W, or grab the laptop version with much lower power consumption, and it would still outperform a T-bird. (I grabbed the first power measurements I could find and assumed people could do the scaling in their head.) Modern CPUs also do the power throttling you describe already.

Snazzy effects

[by Anonymous Coward]

As cool as these Compiz effects are, they should not be forced upon everyone, just made very easy for people to obtain.

Plus, this version never actually booted up because it didn't like my Broadcom 4318.

Re:Snazzy effects

[F-3582 (996772)]

Rest assured, it takes you four mouse-clicks to disable them. Every tried that under Vista?

Re:Snazzy effects

[DAldredge (2353)]

Vista takes 5.

1: Right click on desktop.
2: Select Personalize
3: Select Theme
4: Select Windows Classic
5: Click OK.

Re:

[DAldredge (2353)]

If it took you 5 hours to figure that out then you really do not need to be using a computer...

Re:

[sgtrock (191182)]

In order to prevent active attacks, device drivers are required to poll the underlying hardware every 30ms for digital outputs and every 150 ms for analog ones to ensure that everything appears kosher. This means that even with nothing else happening in the system, a mass of assorted drivers has to wake up thirty times a second just to ensure that... nothing continues to happen

From Peter Gutmann's excellent "A Cost Analysis of Windows Vista Content Protection" [auckland.ac.nz]. This paper should be required reading for any

Re:DRM effects. Re:Snazzy effects

[DAldredge (2353)]

From Wikipedia Criticism of Peter Gutmann's analysis of Vista DRM Peter Gutmann's Vista criticism has come under fire after his speech at the USENIX Security Symposium in August 2007.[3] from George Ou (ZDNet) who challenged Peter Gutmann's claims that Vista Content Protection causes so much additional CPU utilization that it increases power consumption and causes global warming.[4] Gutmann made many of the basic assertions in his paper on Vista content protection but made the more extreme statements at Usenix Boston 2007 as reported by PCWorld.[3] Ou cited data that showed no measurable power differences between 5% and 15% CPU utilization on an Intel E6600 dual-core processor and then cited HD playback performance data from AnandTech which indicated less than 7% total CPU consumption during 1080p VC-1 encoded video playback.[5] Ed Bott challenged some of Peter Gutmann's other claims.[6] Ken Fisher challenged Gutmann's claim that Vista content protection extended beyond commercial content in to user generated content.[7] Gutmann admittedly doesn't run Windows Vista and stated in his paper: "Can others confirm this? I don't run Vista yet, but if this is true then it would seem to disconfirm Microsoft's claims that the content protection doesn't interfere with playback and is only active when premium content is present". This statement has recently been removed from Gutmann's website but an older PDF version the paper with that statement can be found here. George Ou later reported that Gutmann relied on web forum postings for several of his key assertions such as excessive CPU and memory consumption in Vista's Media Foundation Protected Pipeline (mfpmp.exe) and AudioDG (Windows Audio Device Graph Isolation) process. Ou's tests showed that the web forum data Gutmann relied on were not repeatable. Furthermore, CPU utilization was wrongly attributed to mfpmp.exe when in fact it was actually accounting for all the CPU consumption in mfpmp.exe and Windows Media Player 11 combined.

Re:

[PitaBred (632671)]

So... in [3], what Ou is saying is that even if Vista doesn't use more "power", it's still using 3x the CPU resources to do something that actively detracts from a consumer's options in using media? And this is somehow... good for the consumer, wasting resources without giving the owner of the machine a choice in the matter? Just because it doesn't use more power from 5-15% doesn't mean that the difference between 50% and 60% utilization won't change the wattage draw. I think y'all got something a little

Please read Gutmann's work yourself

[sgtrock (191182)]

First, I'll let Gutmann comment on his use of various OSes [auckland.ac.nz]:

This is just Microsoft-bashing.

It's bad-technology bashing. If this had been done by Linus Torvalds, Steve Jobs, Alan Cox, or Theo de Raadt, I'd have said the same thing about it. As far as I'm concerned computers are tools to get a job done and not a platform for religious wars, and if something's bad I'll say so regardless of who's doing it. In fact Vista overall has some really nice new technology and features built into it, it's just this one

Re:

[Runefox (905204)]

I'd prefer a HW-accelerated interface to one that's more or less driven by the CPU. The flashy effects can be turned off if you don't like that, though some of them (desktop cube, "expose", and so on) actually provide some utility.

I'm still very disappointed in any OS that can't wobble its windows.

Boingy boingy. Amusing for idle time.

Other OSes?

[p0tat03 (985078)]

I'd be more interested in seeing how Ubuntu's power consumption stacks up against Windows and MacOS...

Re:Other OSes?

[mpetch (692893)]

Me too, but I'd need to be able actually read the article with Safari on OS/X first.

Re:Other OSes?

[tomee (792877)]

I agree. There also seems to be no info on whether they used the 3d-desktop stuff. I would imagine that that would have a much greater impact on the power consumption, and it would be interesting to see some data on that.

Re:Other OSes?

[Splab (574204)]

In my own experience running Ubuntu 7.04 gives me somewhere between 5,5 hours and 6 hours of battery time. Running windows XP usually gives me around 8 hours of battery time. I use both distributions for development, I code under MS VS (C++) in windows and use VIM under linux for development (also C++).

Gonna check out the new 7.10 and see if I can get nearer to what windows can give me.

How to test the power comsumption of Linux

[by Anonymous Coward]

1. Run the particular distro.
2. Make sure the power cord is long and in the open.
3. Allow a penguin to chew through the cord.
4. Measure the distance the penguin flies after chewing into the cord. This will give you some idea of the power usage.
5. Well, don't let that penguin go to waste! BBQ and teriyaki are great ways to make penguin. Personally, I prefer General Tso's Penguin myself.

Sig Fig nitpick

[by Anonymous Coward]

TFA doesn't specify error bars, which of course makes the results somewhat dubious. They list numbers to two decimal place accuracy (e.g. 48.00), but since all the numbers end in .00, I'm guessing those decimal places are not significant. In other words, the number are only good to within +/- 1 or +/- 2 or something like that. Considering that they are trying to compare numbers that are quite similar (27 to 33), their conclusions may not be reliable.

When comparing numbers, an estimate of the error is crucial. If the difference between two measurements is smaller than the error, then you cannot meaningfully say they are different.

Re:

[DarkMantle (784415)]

Sorry to ruin your nit pick but a quick google search tells me that the SeaSonic PowerAngel [seasonic.com] used in the test has an accurazy of 2%.

So a 2% variance on 33 watts is between 32.33 and 33.66. The 27 would be between 26.46 and 27.54.
of course, this is approxamate I just got what 2% of 33 was and added/subtracted. It's the lazy mans math.

Re:

[volsung (378)]

So that's one piece of the uncertainty on the measurement. The next piece we need is how much the power reading fluctuates while the computer is in a "steady" state. Using my Kill-A-Watt, I've seen short time variations of a few watts on a computer (though it drew more power than a laptop).

Re:

[volsung (378)]

One could roughly estimate the variance by looking at the meter fluctuations while taking the reading, or checking the design accuracy of the meter in the manufacturer's data sheet. You need some kind of estimate if you are going to draw any conclusions (which the authors of TFA were attempting to do).

Does Ubuntu benchmark this kind of thing?

[maubp (303462)]

Does anyone know if Ubuntu actually have machines hooked up to measure this kind of thing in house, looking for regressions in power usage? I gather the OLPC project has thing kind of thing as part of their build system (driven off the code repository), but they are naturally particularly concerned about battery life.

Re:

[LingNoi (1066278)]

I don't think Canonical does much in-house testing. I thought they got the community to do almost everything to do with Ubuntu, apart from employees who live around the globe and commit their code to the servers.

misleading

[wizardforce (1005805)]

very misleading healine. I RTFA and if you look at the nice graph, it actually shows a decrease in power usage since feisty and just about what the prior versions were. AC power consumption idling went from 31 to 29 from feisty to gutsy. while loaded, it went down slightly from 51 [feisty] to 50 [gutsy] the only thing that gutsy was higher in was battery discharge rate idle- it was at 22.26 while feisty was at 21.16. while loaded on battery it went down from 33.51 to 32.21 from feisty to gutsy.

News: More Processing Requires More Power

[reidbold (55120)]

toast, still not free

Well duh!

[pair-a-noyd (594371)]

"more daemons and processes running by default on these modern Ubuntu releases is actually causing an increase in power consumption."

Ya think?

I just install OpenSuse 10.3 on a tower type and a laptop.
The first thing I do is go in and disable a whole slew of bullshit that's enabled by default.
I LOVE Linux but the trend lately has been to BLOAT it up like a new eMachine that's preloaded with 40gigs of bullshit.

What ever happened to minimal? When I installed Suse 9.3 on my Athlon 64 w/1g ram, it ran like a cat with it's ass on fire.
SAME hardware with OpenSuse 10.2 was abysmal. It was sooooo bad that I was just about to give up on it then 10.3 came out.
It's a slight improvement but, damn! They are developing all the new distros with the assumption that everyone is going to run out and buy all new shit. Shades of M$, dare I say??

For the longest time Linux captured and held my heart because it would run so fast on the oldest, worst case hardware.
No more. Wanna run the latest distro? Better put some $$$ back for all new hardware...

Bloat = power drain.

How about getting back to basics and quit focusing on the bling-bling. Linux is NOT windows and it never should be. Quit trying to make it look and act like windows. Quit trying to make it run windows crap. Be happy that it's not windows. I do not want windows compatibility. At all. Ever.

Kill the bloat and pork and watch power consumption go down. Not to mention the old PC's being tossed out into the environment.

Re:

[dvice_null (981029)]

> What ever happened to minimal?

You would like to have a light Linux distribution? Something like this perhaps:

http://www.puppylinux.com/ [puppylinux.com]
http://featherlinux.berlios.de/about.htm [berlios.de]
http://www.damnsmalllinux.org/ [damnsmalllinux.org]

Re:Well duh!

[Simon (S2) (600188)]

There is. http://www.xubuntu.org/ [xubuntu.org]

It is lighter on system requirements and tends to be more efficient than Ubuntu with GNOME or KDE, since it uses the Xfce Desktop environment, which makes it ideal for old or low-end machines, thin-client networks, or for those who would like to get more performance out of their hardware.

Re:Well duh!

[BlueParrot (965239)]

How about getting back to basics and quit focusing on the bling-bling. Linux is NOT windows and it never should be. Quit trying to make it look and act like windows. Quit trying to make it run windows crap. Be happy that it's not windows. I do not want windows compatibility. At all. Ever.

Kill the bloat and pork and watch power consumption go down. Not to mention the old PC's being tossed out into the environment.

Ubuntu certainly isn't windows. That is why you can open the package manager and purge most of the stuff that you find bloated, or use Xubuntu, which is designed to have lower requirements yet still be easy to use. Or if you REALLY want to streamline your system you could install a distro with that purpose, like DSL or Feather Linux. If that is too limited for your needs you could grab a minimal debian install and only install the packages you want.

My point? Different users have different needs. Ubuntu is explicitly targeted and those people who WANT an easy to use GUI and those people who WANT painless support for things they expect to just work. Making an operating system which caters to those users is the main purpose of the Ubuntu project. If your main priority is a streamlined system, then quite frankly you should be looking at something targeted at that rather than complaining about Ubuntu. Besides, it is not as if Ubuntu doesn't run just fine on moderate hardware. I'm using the Gutsy beta on a 5+ year old workstation my dad's job threw out because it was "old" as an example.

Re:Well duh!

[corychristison (951993)]

I've got an older system here... I built it roughly 5 years ago (I think -- I may be over-estimating). It's an AMD Athlon XP 2500+, 1GB OEM RAM, 120GB SATA Hard drive, and an Nvidia FX 5200 video card (I think)...

I used to run SuSE 9.1 and was running it fine for 3 years or so... then came time to try and upgrade to a newer version. Of course this is right around the time that Novell bought SUSE and changed it up a bit. So an easy upgrade was indeed not possible. I decided to try out a few distributions but had a lot of problems finding one that would work fast and I ended up on Gentoo. I know, I know, compile time was a pain in the ass... I decided to go down the XFCE route and use all of the lighter-weight GTK programs... I think I only have one QT program that I actually use installed and it only depends on QT, nothing else.

Xubuntu ran O.K... but not anywhere near as nice as Gentoo is. I think it's not the fact that it was compiled and optimized... I beleive it's because during installation I learned more as I set it up. And I knew what I wanted/needed to run the system. Whereas Ubuntu makes a lot of choices for you, mostly in system services, etc. I have a total of 29 items that start up when I boot. I think only 10-15 of them are actually daemons. Right now I am using 215MB or so of my 1GB of RAM... this is with Firefox (4 tabs), Thunderbird w/Lightning, aMSN, Terminal, Mousepad and a whole slew of items on my panel.

If you want lightweight, make sure you know exactly what is going on your system. And use something like XFCE or Fluxbox versus KDE or Gnome.

Just my two cents. :-)

More features - Same power

[Cryophallion (1129715)]

Actually, I think it is rather impressive that 7.10 (which has eye candy on by default) has slightly less power consumption than 7.04 (no eye candy by default).

In other words, they increased features while decreasing (generally) power consumption. While it seems to be only about 1 or 2 watts lower (excepting battery idle where it is slightly higher), we are only talking 3-5 watts difference over 2.5 years of updates. In fact, it went down 4 watts using ac idle compared to 5.04, which I am sure had far fewer daemons/features.

Some of this may be better code etc. However, I think we should be giving the people who have been doing the coding here major Kudos for doing getting the most out of our computers (whereas MS wants us to quadruple our ram to use eye candy, they are doing it with the same amount of ram standard 4 years ago on a desktop, and keeping power down). I don't even want to think of what Vista must use in power.

AMD64

[Nimey (114278)]

Laptop users may want to stick with 32-bit Ubuntu, since the CONFIG_NO_HZ (tickless kernel) option isn't available in 64-bit kernels yet.

If you're feeling adventurous, patches here: http://www.kernel.org/pub/linux/kernel/people/tglx/hrtimers/ [kernel.org]

I can believe that

[MikeRT (947531)]

I was pretty amazed with the beta of Ubuntu 7.10. I even installed it on my system, but after about 30min-1 hour of use, trackerd was consistently keeping my CPU usage up at least 30%. That's not the fault of the Ubuntu team, as they did not write trackerd, but they really do need to be careful about the daemons that they allow to run in the background on a default installation. I don't know what it is there for, but according to this description [eyrie.org], it doesn't sound like it is something that a vanilla, desktop installation would want on there. The approach to background processes should be the KISS. On a vanilla desktop installation, only the barest set of such thing should be on there.

Re: I can believe that

[Dolda2000 (759023)]

I don't know what it is there for, but according to this description [eyrie.org], it doesn't sound like it is something that a vanilla, desktop installation would want on there.

That's not the trackerd you're looking for, though (for future record: You may want to try dpkg -S /usr/bin/trackerd, followed by dpkg -s $PACKAGENAME to find out what it is). Trackerd in the latest Ubuntu is a desktop search thingie, similar to Spotlight or whatever the Vista thing is called. I'd imagine that the load you were seing after about ½-1 hour of use was that it was still busy indexing your preexisting files. Once it gets past that, it gets quite calm in my admittedly limited experience.

The approach to background processes should be the KISS. On a vanilla desktop installation, only the barest set of such thing should be on there.

If that's what you want, maybe you shouldn't be using Ubuntu?

Re:

[EvilRyry (1025309)]

There is an option to not index while on battery power. I know in earlier alphas that option didn't actually work, hopefully that is fixed by now.

too much crap

[m2943 (1140797)]

There are way too many daemons running on modern Linux systems; it really shouldn't require separate processes for I/O, settings, hardware configuration, every little panel thingy, etc.

Re:

[A beautiful mind (821714)]

So what are the alternatives?

Either you bundle them in one horrid complex mess where maintainability gets inverse squared with any new feature, or you simply don't use the feature at all. With lots of separate daemons you have the option to axe those you don't want/need and the rest are benefiting from a modularized way of looking at things.

Good but...

[suv4x4 (956391)]

I really wish they put Ubuntu next to latest XP, latest Vista and latest OSX (ok I guess they could wait few days for Leopard to get out).

As an XP user, two Ubuntu tests don't give me a clear picture of how this relates to the OS I use right now. I do suspect Ubuntu will have lower power consumption than XP, and for Vista the margin will be pretty wide.

But how much exactly..?

30 Watts? WTF?

[zdzichu (100333)]

There must be something very wrong with tested system or Ubuntu configuration. 30 watts idle consumption is very, very wrong. My Thinkpad z61t idles at 13-14W with Ubuntu 7.04 Feisty Fawn. I expect lower value after upgrading to 7.10. And maybe even lower when I roll my own latest kernel with patches from lesswatts.org. I would be happy to go to 10W on idle, it would match advertised 6.25 hr worktime with 65Wh battery I have.