Computers, electricity, and you

Measuring power use and cost

Electricity consumption is generally measured in kilowatt hours (KWh). A kilowatt is a thousand watts, and that's more electricity than any single device in your home will consume at once. You won't know how much power each device consumes unless you measure it (although refrigerators often have stickers that list their rate of power consumption). So how do you do that?

Although you can use a multimeter to measure power usage, I've found that specialized meters are a better solution. If you think that you're spending far too much on power consumption in your home, the price of the meter is negligible. Though there are a few different meters that specifically measure electricity consumption, I've found the Watts Up Pro to be the most cost-effective.

Next, you need to know what your cost per kilowatt hour is. You can check your utility bill to find out for sure, but for the purpose of this article, I'm going to use the average cost per KWh for my home state of Florida: $0.0731 (a little more than seven cents per KWh). That's about the middle of the energy cost spectrum in America. You can look up your state's average cost per KWh as of the year 2002 at this site, which uses data from the US Department of Energy.

Rigor calefacta

Obviously, refusing to give power to a device means that it will not cost you any money to operate. But you have a computer so that you can use it, not so that you can marvel at how much money you're saving by not using it. You could turn it on only when you need it, but that puts a lot of thermal stress on a machine.

Thermal stress is among the top causes of failure of electronic and electromechanical devices. It's not just heating something up or making it cold that causes damage or a shortened lifespan -- although these stimuli in extremes can be fatal to electronic devices -- it's thermal stress, or the rapid transition from one thermal state to another. In other words, turning a device on and off -- a power cycle. Turn something on and off enough times and eventually it will fail to start.

Most people are already familiar with this concept as it relates to automobiles. We all know that city miles are harder on a car (and on fuel efficiency) than highway miles. In other words, when the car is frequently starting and stopping, it undergoes more stress and uses more fuel. When it is running at a consistent temperature and pace, it becomes more efficient and has a longer life. The same idea applies to all mechanical, electrical, and electromechanical devices.

Your hard drive is the most important internal component of your computer. If it fails, you lose all of your software and data, rendering the computer useless in a home desktop setting (networked computers can use remote storage instead of a hard drive, but that requires a great deal of know-how and at least one other working, networked computer). It's also the part of your system that will probably fail first because of its reliance on electric motors. The drive platters have to spin at speeds as high as 10,000 rotations per minute, and the motor that controls the read/write heads has to skim over the surfaces of the platters with exact precision. This is a recipe for failure because of the high amount of friction, but currently it's the only way to quickly, cheaply, and reliably store large amounts of data in a computer system.

So do you save money by turning the system off when you're done with it, or do you leave it on all the time to maximize the life of the machine? The answer depends on how much money you're spending on electricity for the computer, and how often you use it. If the computer is very important to you and you use it often, it's probably a good idea to leave it on most of the time. On the other hand, if your energy bill is unacceptable, you might have to sacrifice the computer's longevity for short-term financial stability.

If you're looking to save some money on your energy bill and are ready to buy a new computer or new computer parts, what choices will yield the best energy savings? I measured a variety of scenarios over a period of fifteen minutes, starting from five seconds after the power switch was activated (this delay avoids recording the huge power spike that a computer experiences when it is initially powered on). I recorded the watt hours consumed in that timespan and projected them out to a one hour period, the average monthly kilowatt hour (KWh) usage, the minimum and maximum wattage, calculated the cost based on the above-mentioned cost per KWh, and organized the data into the following comparisons:

Dual Opteron or Dual G5?

It's the original personal computer argument: Apple or Intel/AMD? I hardly think that power consumption will sway any Apple enthusiast's opinion, but for the sake of measurement, let's see which system is hungrier.

The first system is an Apple PowerMac with two G5 processors at 2.0GHz, 5.5GB RAM, two 250GB 7200RPM SATA hard drives, a wireless network card, an ATI Radeon 9800 video card, and a 20" LCD screen. The operating system is OS X 10.4.3. The second system is a Sun Java Workstation w2100z with two AMD Opteron 252 processors, 4GB RAM, a 72GB 10000RPM SCSI-360 and a 180GB 7200RPM SATA drive, an Nvidia Quadro FX3000 video card, and a separately powered (not measured for this comparison) Samsung SyncMaster 997DF 19" CRT display. The operating system is Gentoo Linux for AMD64.

It's hard to find two machines that can compare "fairly," and I have no doubt that the "losing" side of this comparison will scream and cry about fairness no matter what machines are chosen. The only major point of difference that I could see was the LCD monitor on the Mac, which is powered through the data cable. The tests below show that a 17" Apple LCD adds about 8.8Wh to the numbers, so a 20" will draw somewhat more. Anyway, here are the results:

Dual G5 or Dual G4?

Now that you know how much power a high-end Dual PowerMac G5 uses, let's compare it with its predecessor. The system in question is an Apple PowerMac with two 800MHz G4 processors, the stock DVD writer and 80GB hard drive, 384MB RAM in two modules, an Nvidia GeForce2 video card, and a 17" LCD monitor (powered through the data cable, no USB devices attached).

Pentium D or Athlon 64 X2?

Word around the Internet is, Intel's processors draw a lot more power than their AMD equivalents. I don't know if the Pentium D 820 is equivalent in performance to the AMD Athlon 64 X2 3800+ because I did not measure speed, but they are of the same technological generation and fairly close in price, so as far as I'm concerned they are close enough to compare power consumption.

Each system used an Antec TrueBlue 480 power supply; 1GB of either DDR2-533 or DDR400 RAM in two modules; one Seagate SATA-V 180GB hard drive; a Matrox G550 1X PCIe video card; and a Lite-On 52X CDRW/DVD-ROM drive. The Intel machine was based on an Asus P5WD2 motherboard, and the AMD machine used an Asus A8N-E.

The real challenge was finding an operating system that worked on both machines. Well, actually the challenge was getting anything to work properly with both motherboards and the video card. Windows XP had problems, nearly every GNU/Linux distro had problems, and the *BSDs didn't have drivers that fully supported the graphics card. I ended up settling on Knoppix version 3.9 for x86. It's not the best choice for accurately measuring power usage because it runs from a CD, and that means that the CDRW drive will suck in some extra power. In other words, the numbers are inflated by roughly 5Wh.

Clearly the Intel machine draws a lot more power than the AMD machine, and they are using as identical a hardware configuration as possible. The Intel machine could cost you more than $36 more per year to run (or more, if your cost per KWh is more than 7.31 cents). Multiply that times an office or school full of machines and you've got a lot of wasted money flowing through your power strips.

Also note that the Apple G5 machine above draws almost two and a half times as much electricity as the Athlon 64 X2. Of course the numbers are a little skewed for a direct comparison between the two machines, because you'd have to load up the X2 with a lot more RAM, another hard drive, a more powerful video card, and add in a 20" LCD monitor. But compare it to the G4 PowerMac, which has one hard drive, a lot less RAM, and a more comparable video card. It still uses more than 50% more electricity than the Athlon 64 X2 (and slightly more than the power-hogging Pentium D) and is considerably weaker in terms of computing power. Minor variables aside, the difference is striking, especially considering the hardware costs involved.

LCD or CRT?

Using the aforementioned Sun Java Workstation, I measured the power used by a Samsung SyncMaster 997DF (among the better 19" CRTs on the market), and with the G4 PowerMac I tested an Apple 17" LCD that was purchased with it. The Apple screen gets its power through the DVI connector, so the power measurement was a little more difficult than usual -- I had to measure the power with the LCD disconnected and then subtract that from the whole system's readings.

Secondly, the CRT's power readings fluctuated wildly. When there is a lot of white on the screen, the CRT uses considerably more power than when the screen is mostly black. In a blank, hardware-accelerated GNU/Linux virtual terminal, the power usage was around 50W. Switching into GNOME with a typical light-colored theme made it jump more than 50% -- up to about 77W. The LCD monitor used pretty much the same amount of power no matter what the screen was showing, so it had a more consistent reading.

So that big CRT on your desk is costing you more than twice as much to operate as a comparable LCD screen would. A few years ago that cost difference did not justify the increased cost of good LCDs, but LCD prices have come down considerably over the past two years.

Linux or Windows?

I tested five OSes on the same Pentium D-based machine: Windows XP Professional, SUSE Linux 10/x86, OpenBSD 3.8/AMD64, Mandriva Linux 2006 PowerPack Edition for AMD64/EM64T, and the same numbers I recorded before with Knoppix 3.9 for x86. Where applicable, all of them were installed with default options and immediately updated with the latest patches and service packs at the time of testing. Once the system was updated and rebooted, I used it normally until the fifteen minute test period was over; usually this involved playing solitaire, browsing the Web, and in OpenBSD's case, downloading the Vim package and typing part of a short story into a text file. This ensured that no power-saving functions would kick in, and it also portrayed a more accurate power draw on the system than just letting it sit idle. While the non-standard test procedure leaves room for variance, I found only a difference of less than 1Wh between test iterations.

Not much difference between Windows XP and GNU/Linux, except for the min/max fluctuations. Initial readings for Mandriva showed much higher numbers, so I contacted some programmers at Mandriva and they suggested that the problem may be with an ACPI module, or the Kat search tool may be indexing in the background. I believe it was the latter; I reinstalled the whole OS from scratch, applied all patches, let the system stay on all night, then rebooted and tested. The numbers from that test are the ones you see above. What this may mean is that increased or sustained hard drive activity can significantly affect power usage, even when there is only one hard drive in the system.

OpenBSD almost falls within the margin of error. Since the video card wasn't doing much, I assume that this was the cause of the lower power usage. Maybe operating systems don't mean much when it comes to power consumption.

Everything the light touches

So just how much power is my whole workbench sucking down? This test combines the Sun Java Workstation w2100z, Samsung SyncMaster 997DF monitor, Altec-Lansing ATP3 speakers, and a Logitech MX1000 laser mouse.

There were significant fluctuations depending on what the computer was doing. Once I reached the text mode login prompt, the average meter reading was around 317 watts. When I started X.org and loaded GNOME 2.10, that number shot up to over 340. Then I started Unreal Tournament 2004 and watched the meter go over 400 watts. I played the game for about three minutes, then quit and did normal desktop work -- email, Web, writing, etc. until the test period was up.

Conclusions

If you buy computer parts and peripherals with power consumption savings in mind, you can make a significant difference in your energy bill. By choosing an LCD over a CRT monitor, an AMD-based machine rather than an Intel or Apple computer, or a dual-core machine over a dual-processor machine, you can lower your electricity costs while not necessarily reducing computing performance. And if money's really tight, you may want to think about switching to GNU/Linux or *BSD and working mostly from the command line. I doubt many people would take that route just to save a few dollars on electricity on a desktop computer, but consider the implications for laptop systems. How much battery life does your notebook computer have? Working from the command line could greatly increase the length of time you can use your system away from a wall socket. Anyone who has had a layover at New York's JFK airport and searched for the hidden power outlet (no, I will not tell you where it is) knows the value of maximizing laptop battery life.

What about servers and workstations? Cost savings aside, in places like California where there are regional limits on how much power a business can consume based on the size of the building, an IT manager must consider power usage when building a server farm. Instead of big, power-hungry Intel Xeon machines, a company could increase the number of servers it can operate by choosing a more efficient dual-core Athlon 64 X2-based configuration. Or in offices that require a large number of workstation or desktop machines, abandoning Apple in favor of drastically more efficient AMD64/EM64T-based computers could make a big difference.

I mentioned above that the difference between AMD and Intel in the test scenario meant an extra $36 per year for Intel machines. That's per computer, so if you have a school in Florida with two 30-machine labs, that adds up to $2160 of extra electricity per year, just by choosing Intel over AMD. Instead of upgrading to G5s, switching from PowerMac G4 dual-CPU machines to more powerful Athlon 64 X2-based computers could mean a savings of $2325 over the course of a year, based on the data above. That doesn't count the extra savings beyond the higher cost of operating Apple hardware.

If you're gung-ho about saving money on electricity now that you've read this article, remember that every power cycle brings your electronic devices closer to death. There's no way to "win," really -- one way or the other you're spending money. The question is how much and how often.