Profiling the Power Usage of a Desktop PC
Reducing the power usage of a desktop computer can bring about a number of benefits. Whether your goal is to save money on your power bill, reduce your carbon footprint or eliminate unwanted heat and noise from your office, a bit of effort can produce a more power-efficient computer. Effort spent reducing power can have an even larger effect on servers and other machines that run 24 hours a day compared to machines that are only on during work hours. This work was done on a nearly ten year old PC, but the process still applies to more modern hardware.
The test setup consisted of an opened-up desktop PC, a P3 International Kill-a-watt meter and a collection of peripheral cards and disk drives. The Kill-a-watt has a 1W resolution, if a reading alternated between 2 values such as 8 and 9 Watts, the estimated value was called 8.5 Watts. Some of the measurements made were small enough that they were "in the noise". Other variables included devices with inconsistent power usage and inconsistent line voltage. The resulting measurements were actual power used by the power supply, this may vary from the DC power used by the tested components. Lastly, the Kill-a-watt meter also shows power factor; a fairly consistent value of 0.67 was read.
![[PC Power Test]](https://static.lwn.net/images/ns/PCPowerTest-small.png)
The tests were performed on the machine while it was in a number of different software states. Many of the tests were done while at the BIOS prompt, disk drive and network adapter tests were done while the machine was running Linux (Ubuntu 8.10). Power consumed by external devices such as the LCD video monitor and amplified speakers was not taken into account. When a peripheral such as a disk drive was removed for a test, the drive was disconnected from power and the interface cable was removed to eliminate possible power consumption by bus termination resistors.
The tested computer used a fairly old, but still adequate Asus A7V333 motherboard with an AMD Athlon 1700 processor clocked at 1466 Mhz. The RAID option was not present on the motherboard. A pair of 256MB PC2700 DIMMs were used for the memory. The power supply was a 300W Antec PP-303X. Initially, the machine was loaded down with two hard drives, both CDR and DVD-RW drives, a floppy drive, an AGP video card with an ATI Radeon 8500 GPU, and both wired and wireless 802.11 networking cards.
The machine was shut down, all of the PCI and AGP cards were removed and the disks were disconnected. The first power test involved the PC2700 memory DIMMs. With no memory, power consumption was 72 Watts. Adding one DIMM caused the power to drop to 67 Watts. Your author guesses that with no memory, the CPU runs in some kind of power-consuming loop. Interestingly, the two DIMMs had significantly different power usage. The Kensington Value Ram with Hynix chips caused the machine to use 73 Watts versus 67 Watts with the generic Chinese RAM with unbranded chips. With both DIMMS installed, power consumption as 75 Watts. We can deduce that the Kensington RAM used 8 Watts while the Chinese RAM used 2 Watts. Sufficient RAM is critical for good system performance, the brand seems to be significant in the area of power usage. Tests with additional brands of memory seem to be in order.
Fans consume a fair amount of power. A quick unplugging of the noisy CPU fan caused the power to go from 75 Watts to 72 Watts, the CPU would melt down without this 3 Watt component, so it was left in place. It may be possible to find a more efficient CPU fan. The case had a front-mounted "push fan". This consumed around 2 Watts of power. The power supply's built-in fan provides plenty of air circulation so the front fan was disconnected. This also made the machine a bit quieter.
The floppy drive is virtually useless now that 4GB USB memory sticks can be purchased for under $10. The floppy drive consumes about one half Watt of power, so the savings are small. But big savings can come from many small cuts, so the device was left unplugged. The Asus CD-S500/A CDR drive was tested, it consumed about 1 Watt of power. The Sony CRX320E DVD-RW drive was tested, it consumed about 2 Watts of power. Most people can get by with a single removable media drive, or none at all. The DVD-RW drive would be the obvious choice for a single-drive system. If one can put up with the occasional inconvenience of rebooting, it should be possible to put a DPDT power switch on the back of the machine to allow shutting off the +5V and +12V lines to the removable media drive. All together, the floppy and two optical drives consumed around 3.5W when idle.
The Radeon 8200 video card was somewhat of a power hog, it consumed around 8 Watts of power with no built-in fan. A lower performance ATI-S3 AGP video card consumed 4 Watts. If high performance video operation is not critical, example: running Google Earth, the S3 card should be sufficient. As with sufficient memory, this sacrifice may not be worth the power savings.
The next part of the power test involved the fixed disk drives. The main boot device was a Western Digital WD600 60GB PATA disk. It consumed about 7 Watts of power at the BIOS prompt, power went up by about 5 Watts when the system was running Linux and the drive was active. Some of this power is likely being consumed by the CPU and memory and some is used to power the disk's head actuator motor. An auxiliary Western Digital WD2500 250GB SATA drive and associated SATA PCI adapter card consumed around 9 Watts of power when idle and also about 5 watts more when active. Interestingly, as the machine was more heavily loaded with drives and peripherals, system usage became less of a variable to overall power consumption. Hard drives are one of the more power hungry devices in a system, putting all of your data on a single drive is a good way to save power.
A generic-brand 10/100 Ethernet controller with an Intel chip consumed about 1 Watt of power at the BIOS level. Running Linux and moving a lot of data across the card caused the power consumption to jump by about 8 Watts, as with the disk drive test, a lot of that increase is likely caused by CPU and memory use. A Hawking Technology HWP54G 802.11 wireless Ethernet card also consumed about 1 Watt when idle and a few watts more when busy.
The fully loaded system with 512MB of RAM, two hard drives, two optical drives, two network adapters, the Radeon video the floppy disk drive and the front fan consumed about 108 Watts of power when idle and a similar amount when busy. When the machine was stripped down to one hard drive, no optical or floppy drives, the lower performance S3 video card and no front fan, its power dropped to 80 Watts idle and 88 Watts when busy, or between 74 and 81 percent of the original power consumption. This is enough of a reduction in power usage to justify the effort of testing.
Don't forget that even when it is completely powered down, the computer may still act as a phantom load, this system consumed a full 3 Watts when it was off. An easy remedy to that problem is to route the power plugs for the CPU, video monitor and speaker through a switched power strip.
Posted Dec 18, 2008 5:00 UTC (Thu)
by eru (subscriber, #2753)
[Link]
Posted Dec 18, 2008 5:03 UTC (Thu)
by bronson (subscriber, #4806)
[Link] (2 responses)
Posted Dec 18, 2008 17:15 UTC (Thu)
by cook (subscriber, #4)
[Link] (1 responses)
Posted Dec 18, 2008 18:32 UTC (Thu)
by knan (subscriber, #3940)
[Link]
Posted Dec 18, 2008 6:57 UTC (Thu)
by kev009 (guest, #43906)
[Link]
That was a 2x PIII server with no X. Basically I saw a free 1-2VA drop.
Posted Dec 18, 2008 7:10 UTC (Thu)
by dlang (guest, #313)
[Link] (1 responses)
Posted Dec 18, 2008 17:17 UTC (Thu)
by cook (subscriber, #4)
[Link]
Posted Dec 18, 2008 11:27 UTC (Thu)
by stevan (guest, #4342)
[Link] (6 responses)
Stevan
Posted Dec 18, 2008 15:21 UTC (Thu)
by deleteme (guest, #49633)
[Link] (5 responses)
Posted Dec 18, 2008 15:22 UTC (Thu)
by deleteme (guest, #49633)
[Link] (3 responses)
Posted Dec 18, 2008 15:31 UTC (Thu)
by deleteme (guest, #49633)
[Link]
Posted Dec 20, 2008 2:15 UTC (Sat)
by branden (guest, #7029)
[Link] (1 responses)
Posted Dec 20, 2008 13:50 UTC (Sat)
by nix (subscriber, #2304)
[Link]
(Lithium-Iron-Plutonium is more practical, although actual designs omit
Posted Dec 20, 2008 22:41 UTC (Sat)
by man_ls (guest, #15091)
[Link]
Posted Dec 18, 2008 11:44 UTC (Thu)
by ms (subscriber, #41272)
[Link] (3 responses)
PSUs are vastly more efficient these days than they used to be. I suspect that if you got a good, high quality PSU and put it in there you would find your power usage reduce further. Both anandtech and silentpcreview do very good PSU reviews.
Posted Dec 18, 2008 13:49 UTC (Thu)
by nettings (subscriber, #429)
[Link] (2 responses)
imho an important part of power tuning is to get a modern psu with good compensation (something like "80+" seems to be the marketing phrase - iiuc it implies a power factor of 0.8 or better) and high efficiency at low power (which is where most offers are sadly lacking...).
Posted Dec 18, 2008 16:29 UTC (Thu)
by ms (subscriber, #41272)
[Link]
Again though, I'm really just repeating material from silentpcreview.
Posted Dec 27, 2008 17:42 UTC (Sat)
by anton (subscriber, #25547)
[Link]
Posted Dec 18, 2008 14:38 UTC (Thu)
by shapr (subscriber, #9077)
[Link] (5 responses)
On a more useful note, I wasn't aware of phantom power.
I also have a cable modem and a wireless access point, what's the average power required by those?
This is useful, thanks!
Posted Dec 18, 2008 15:35 UTC (Thu)
by deleteme (guest, #49633)
[Link] (3 responses)
Posted Dec 18, 2008 17:30 UTC (Thu)
by cook (subscriber, #4)
[Link] (2 responses)
There are a lot of urban legends regarding power usage.
Posted Dec 21, 2008 2:15 UTC (Sun)
by deleteme (guest, #49633)
[Link]
Posted Dec 30, 2008 15:08 UTC (Tue)
by timl (guest, #33836)
[Link]
The thing is that this model apparently has a design defect: when taken completely off power for a while, it will often lose some settings regarding colour calibration. A standard auto-setup won't help, a trip into the service mode menu is in order to reset the contents of the eeprom.
The kicker is that these settings only influence the analog input, so it's no concern for me: I use a DVI link :)
And yes, while switching devices on and off is in general a heavy burden for them (there's a reason incandescent lights usually fail when being switched on!), it often doesn't outweigh the wear and tear of being left on continuously in my experience.
Posted Dec 19, 2008 13:13 UTC (Fri)
by ssam (guest, #46587)
[Link]
turning off wake on lan can reduce the phantom power draw though, as it requires the network card to be alive enough to listen for the wake packet.
Posted Dec 19, 2008 10:56 UTC (Fri)
by NAR (subscriber, #1313)
[Link]
Posted Dec 20, 2008 15:49 UTC (Sat)
by ranmachan (guest, #21283)
[Link]
If you can afford the higher prices, go with notebook disk drives to save power. Or use on of the recent 'Green power' offerings by some manufacturers. In general, keeping the disk spinning is where most of the power goes for an idle disk. And notebook disks are spinning slower and have less mass.
As someone else said, newer 80+ power supplies are more efficient and all currently available power supplies should have power factor compensation. But the latter has basically nothing to do with efficiency, only with the type of load. Ohmic loads have a power factor of 1.0 and that's generally desirable as far as the power network itself is concerned.
My current server uses about 50W idle running Linux.
Next time I'll go for integrated graphics, I read that AMD chipsets have an especially good power/performance ratio.
Posted Dec 20, 2008 23:03 UTC (Sat)
by man_ls (guest, #15091)
[Link] (2 responses)
The decTOP came into service when my NSLU2 died an ignominious death. The downside as a server is the Ethernet<->USB adapter, which is slow and unreliable. Moving big files to and fro is a pain, but you get patient.
Overall my power consumption has gone down maybe 70%, and that wasn't even my main objective; I just wanted to do away with noisy machines. And it has worked; now I know that LCD monitors do make noise. Next time I might try an Eee Box, although I really like the solid state drive.
Posted Dec 21, 2008 2:22 UTC (Sun)
by deleteme (guest, #49633)
[Link] (1 responses)
Posted Dec 21, 2008 10:20 UTC (Sun)
by man_ls (guest, #15091)
[Link]
With the Debian repository come the improved ACPI modules that give complete control over the fan. With a speed of 30% the fan is barely audible and yet temperatures rarely rise above 60 C. These three lines at startup do the trick:
Posted Dec 22, 2008 0:02 UTC (Mon)
by Kamilion (subscriber, #42576)
[Link]
Intel Little Falls 2 boards are pretty cheap (~$80/USD)and fast (1.6Ghz) with their dual core Atom 330s.
Add $20 for 2GB of DDR2,
Combine that with some cheap 8GB SDcards (~$20ea) and a PhotoFast CR-9000 6-Slot SDHC TF to 2.5" SATA II Flash Disk Adapter/Converter (192GB Max) for $100.
For $200 and some sdcards It's very quick and sips power!
Cheap, plays HD, gigabit lan, and you could easily drop the SD Array and boot it from a network. The next revision that drops the 35W GMA950 graphics should be even lighter on the power. It's a shame they paired an 8Watt dual core chip with a 35W desktop chipset. Once the Poulsbo chip gets dropped, we'll have a much more lightweight system.
(Disclaimer: I am a hardware geek.)
Posted Dec 27, 2008 13:03 UTC (Sat)
by muwlgr (guest, #35359)
[Link]
The same is to be said about rotating-media storage. What requires electric motors for operation, is not energy-efficient in computing sense. Unfortunately, current consumer-level solid-state storage still does not give favorable rates of power per capacity or power per transfer speed (only better random access times, mostly). As well, determine some useful stand-by period for your rotating disks and configure it by hdparm -S.
And, as mentioned in the article, stand-by power consumption is equally important, as current systems are often run always-on or at least always-plugged. Air conditioning is also a considerable expense in industrial settings so it is very worth saving.
So, if your "netbook" has cooling fans and/or rotating disk storage, it's not really a netbook.
In my main system, I have ASUS M3N/H-HDMI mobo, 2x1GB DDR2-800 RAM (Kingston KVR), low-watt AMD 4450e CPU (controlled by powernowd), and two Samsung HD753LJ HDDs, as well as a floppy disk, and Lite-On SOHW-832S DVD-RW, and 3 RTL-8139 Ethernet cards plugged in. The PSU is FSP ATX400-PNF. The system is connected to APC SmartUPS-1000VA (together with some little boxes, like ADSL modem, wireless AP, etc.), and its load capacity read by apcupsd is 8.3% with active disks, and 6.2% with disks in stand-by. Don't know what exactly these numbers mean, but with full scale of 1000VA I think that's not bad.
Posted Dec 27, 2008 18:03 UTC (Sat)
by anton (subscriber, #25547)
[Link]
Posted Dec 27, 2008 21:50 UTC (Sat)
by endecotp (guest, #36428)
[Link]
You can include the standby supply in your measurements and see where your 3W of "off" power go.
I have seen some kil-o-watt type devices with a 0.1 W resolution, so if anyone is considering buying one it's worth shopping around.
Posted Dec 29, 2008 12:55 UTC (Mon)
by daenzer (subscriber, #7050)
[Link]
Also, what's an 'ATI-S3' card? :) Assuming it's really an S3 card, you might have a hard time finding Linux drivers for it that actually run Google Earth.
BTW, I recently read in the German magazine c't that some cheap wattmeters can show much too high values because they just multiply medium current by medium voltage rather than integrating the product of the actual values over time. Not sure if this applies to the Kill-a-watt though.
For additional experiments, try measuring watts with different desktop software! I did some informal measurements about a year ago with my home machine, and found that the difference between KDE and a spartan IceWM desktop was about 10W when the user was doing nothing (however, even the lower load was about 220W in this power hog with 3.4 Ghz Pentium D). I guess with KDE it was running more background tasks that kept waking up the processor.
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Prior to the test, the motherboard's BIOS settings were reset to
the factory defaults so the memory settings should have been identical.
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
The removable drive tests were all based on idle power with no
media. As with any disk, spinning it up and accessing data should
increase the power usage.
Profiling the Power Usage of a Desktop PC
Yes I'm off the grid too, so this is very interesting though I've been looking for two things; a very fast/watt server, and a normal usage laptop. I haven't been able to look for a good calculation server though, still just using a more powerfull laptop for that.
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Wow I found this on their forum: OLPC Power usage is 2W-7W
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
the lithium and iron and go in for the 'big lump of plutonium on the end
of a long pole'.)
I'm no Apple fanboy these days, but the mac mini shows excellent power consumption. The new Core Duo machines behave even better when idle (23 W) than the old PowerPC thingies (30 W), but they peak out a bit higher (110 W vs 85 W). Plus, it's even quieter than many laptops and takes very little space.
May I interest the gentleman
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
phantom loads will burn lots of energy on the grid and waste copper, even if it is not (yet) billed to the consumer (at least in germany).
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
imho an important part of power tuning is to get a modern
psu with good compensation (something like "80+" seems to be the
marketing phrase - iiuc it implies a power factor of 0.8 or better)
"80+" means an efficiency of >80% (at various load points). If you
want a good power factor, look for "active PFC"; that will typically
give you a power factor >0.95 at turned-on loads. To optimize
turned-off power consumption, just electrically disconnect the machine
when off; otherwise you will typically see 5-10W of consumption with
pretty abysmal power factors.
Profiling the Power Usage of a Desktop PC
My desktop is always on, I want to be able to reach it with ssh, but this inspires me to put my three monitors and sundry other devices onto their own powerstrip so I can really turn them off when I'm not at home.
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Please elaborate, is this an issue with the video mode?
In a previous job, I was in charge of repairing CRT monitors.
I discovered that shutting off all of the monitors after hours
made them last *much* longer, they would also stay brighter and more focused.
The myth was that the temperature cycling would hurt the circuitry,
I found that constant baking was a much worse problem. At 100+ Watts
each and 30 or so monitors, the power savings was significant.
Reducing the air conditioning load also saves a huge amount of power.
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
Profiling the Power Usage of a Desktop PC
That's with a 2GHz Athlon64 (idling at 1GHz and reduced voltage thanks to cpufreq), two 500GB 7200RPM 3.5" hard drives and an old PCI graphics card.
And when I upgrade the raid disks I'll go for 'green' hard disks (or maybe notebook disks).
I have gone to an extreme position: a decTOP for server (9 W idle) and an Asus Eee 700 for desktop. That is right, the one with 2 GB solid state (disk?) drive and a 570 MHz CPU; it is not even the revised 701 with the 630 MHz chip which can be overclocked. And still it works like a charm for normal tasks like web browsing or document writing in Lyx. For CPU-hungry tasks... well, I don't do those any more. With an external monitor, keyboard and mouse the Eee is the perfect desktop machine: small, efficient and silent.
Less hungry machines
Less hungry machines
The default fan setting of 40% is not so loud, but is certainly quite annoying. Luckily the Eee has a huge and loyal fanbase that has tried every crazy mod, and Debian pages are about the best. So I tried removing the little fan altogether: the temperature rose about 10 C to 70 C and after a couple of hours the SD card started to malfunction.
Less hungry machines
modprobe eee
echo 1 > /proc/eee/fan_manual
echo 30 > /proc/eee/fan_speed
Profiling the Power Usage of a Desktop PC
http://www.newegg.com/Product/Product.aspx?Item=N82E16813...
http://www.newegg.com/Product/Product.aspx?Item=N82E16820...
http://www.dealextreme.com/details.dx/sku.17790
With a total of 6 Class 6 cards, you get pretty respectable data rates of 60-80MBytes/sec and 48GB! When 16 or 32GB cards drop in price to ~$20, this SSD expands pretty cheaply. It's only as good as you make it, but it does work remarkably well even when fed crap like mismatched speeds of 1GB & 2GB cards (The contents of the junk drawer).
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i...
Profiling the Power Usage of a Desktop PC
Look also at temperature sensors in your hard drives. In my practice I have often seen, e.g., Seagate drives running hotter that Samsung (with other features being equal). As well, in some reviews of the new solid-state drives, their temperature was measured hotter than that of some traditional rotating-magnetic ones.
Use fan control features of your system extensively. The slower is the fan, the quieter is overall system operation and the less power the fan takes by itself. Most modern ASUS mobos have this 'Q-Fan' feature which could even stop your fan completely when the sensed temp. is low enough.
Try to have as much your devices built-in on-board as possible. And use their power management facilities. Even if 3D GUI accelerator or Gigabit Ethernet are on-board, they could still draw quite a lot power when their PM features are not used well.
Also, be picky while choosing your PSU. Its "power factor" is really important (not specified, but really measured one).
I have also collected some data about computer
power consumption. It does not profile individual components as
much, but gives data for a wider variety of machines.
Profiling the Power Usage of a Desktop PC
Try measuring the DC currents
GPUs