For the record, it is possible to design a power supply that's efficient at low load levels. But it's far from common.

http://www.silentpcreview.com/article263-page4.html

The Fortron Zen is pretty amazing; 77% efficiency at 52w, going all the way up to 88% efficiency at 300w.

14.8 cents/kwh? I would love to get electricity that cheap. Over here in Germany we pay 0.17 € cents/kwh (which is 0.22 $-cents). And that's already pretty low...

Sorry, I meant 0.17 €, not € cents. And also 0.22 US-$, not cents.

I'm not running it, but I've heard anecdotally that Vista (with it's Aero Glass and other graphical hoopla) IS running video cards, etc. at a much higher level at "idle" than people were seeing under XP.

This doesn't in any way negate what you're describing, but I think it might be reasonable to tweak your calcuations to account for the fact that maybe Google's problems just aren't our problems YET.

14.28 cents is expensive, huh? Sheesh, you Americans don't know you're born! Next thing, you'll be whinging about the price of "gas" ;-)

Insanely expensive power rates in your area?

I believe that you just proved the point that electric is way too cheap in California. Where is the incentive to buy/use more efficient power supplies at those (cheap) rates? As you point out the technology is out there, but nobody will buy it if it isn't priced competitively and has good amortization. As Jonas points out power is much more expensive in Europe, but then again consumers take power usage into consideration when purchasing appliances which overall has a positive side effect for everybody's future.

As a side note Google is using so much electricity that Solar energy starts to become economical for them.
http://www.greenbiz.com/news/news_third.cfm?NewsID=34136

I've been considering buying the PicoPSU ( http://www.silentpcreview.com/article601-page1.html ) for my already low-power server (no video at all! 5V fans, runs at 300mhz at idle), but since it's been over two years since I built it and a year since I pulled the video card, I haven't bothered; it can't be using more than 70-80W as-is.

As a side note, I love how exhaustively researched all of your articles are, Jeff. =D

The last calculation here is actually only correct in your situation where you already have a power supply.

If someone is looking into buying new hardware anyway, your calculation shows that a better but more expensive powersupply can be cheaper after 2 years, even if it cost 30 - 50 dollar more.

BTW: Keep up the good work.

Here's an outstanding ppt from James Hamilton outlining what commodity data centers will look like in the no so distant future.

http://research.microsoft.com/~jamesrh/TalksAndPapers/JamesRH_CommodityDataCenterDesign.ppt

Of course, the question could also be whether looking for a financial incentive is the way to go on powersaving. Green computing is something we not only do for our wallets, but also for our children?

"the greater public good"

Coding Tau?

You can save money now, or you can save later. It's all preference.

Good as always Jeff, nice reading. :)

#jeffpage:
{
background-color : #00FF00;
}

Anyway, as Jonas said your prices are very low. Here in Denmark we pay around 0.33$/kWh. Gas prices has just exploded, around 11DKR/l, should be around 7.45$/galon as far as I can calculate.

Hi,
Just checking, but did you calculate your efficency values in that table by dividing one voltage by the other...that's not how you do it, that would only be the ratio of the different voltages.

Or is it just coincedence that the percentages come out that way?

Best regards
Steve

I don't get it. You said prices are "insanely" high but still you say it would take years to compensate the initial costs of a better power supply. Isn't something wrong there?

If energy prices were higher, the better power supply would amortize in the first year and more people would make the switch. So energy is TOO CHEAP. The first priority should always be energy savings, not your personal money issues...

But still a great post, as always :)

Sorry, scratch that...I misread it while scanning...

S.

blah blah blah, there so innovative.

That's why they chose NC as their datacenter site. Tax breaks, a couple of COAL power plants, and the legislature cheaply brought.

Google are locating their data centers next to power substations (like dams) where the direct cost is a lot lower, and then kitting out their roofing with solar technology. I'd say they're very aware of the cost implications and reacting to it sensibly.

Thanks for the story. One small correction: The "Holze" referenced briefly above (the guy personally offended by inefficient power supply) is actually "Holzle" — or more precisely, Urs Hölzle, Google's Senior VP of Operations.

"Idle power consumption for a typical desktop PC ranges between 120 and 150 watts. Thus, the real challenge is to deliver 90%+ efficiency at typical idle power consumption levels-- 120-150 watts."

No, the real challenge is to drop power consumption levels on idle to something *much* lower.

On idle, putting the CPU into a lower energy mode, dropping it's speed to the 100MHz range, spinning down disks, powering off CD drives, switching off displays, and making displays and other peripherals draw *much* less power in standby modes, *these* are the real challenges that will give real savings at home.

They might not work for google as google probably doesn't have its boxen idling very much. But for the home, this will give much better benefits.

Computer power efficiency is definitely heating up as a green topic.

1) Even though we can do a lot to make computing more efficient, it's still a far more efficient industry ($ produced/W) than anything else going. When Google pumps out great maps for people (who then avoid getting lost) or people shop online (avoiding trips to the store), the net watts are pretty negative.

2) Computing in the cloud also has strong energy implications: http://profitdesk.com/content/2006/10/13/saving-it-costs-by-banking-in-the-cloud/

My biggest issue for the last 5 years or so has been the requirement for a high end 400 watt power supply just to be able to get my desktop at home to boot with 2 optical drives, 2 hard drives, and a high end (for 5 years ago ;p) video card. With a less efficient 350 watt power supply the drives would all spin up during the boot process and the system would promptly reset as something was starved for power. That alone was reason enough for me not to turn the computer off unless I really needed to, as I can't imagine the power load being high enough at idle with the power saving features enabled to counter the cost of turning the thing on whenever I need it.

Of course, the surprising thing (and this should be to most people) is that power consumption has continued to increase, as most of the chip makers (in both video and CPU chips) have touted all of their new power savings, but it seems more likely that they're just making the chips efficient enough that they can actually be used in the home, rather than efficient enough to use less power overall.

Cisco's newest labs have *most* devices running off DC (48V). They don't have the overhead (heat) of AC-DC conversion in every device.

Do not generalize high electricity prices in Europe, in France it is 0,07€/kWh. But the energy sector is not yet deregulated.

Maybe we need hybrid power supplies, like hybrid cars.

It's a bit of a joke, but in all seriousness, are notebooks more efficient because of the battery? I imagine they would have to be. For home PC's (not servers), a power supply that runs off of battery when the computer is idle might increase overall efficiency, just like hybrid cars remove polution caused by idling cars.

Of course, if you leave your charger plugged in it could blow the whole equation (Nokia recently warned that 2/3 of all cell phone power usage is from adapter bricks being plugged in while not charging).

Hardware isn't getting effectively free. That's an absolutely pointless math.

Although technology creates more computing power it also creates the demand for it. I could work, play and live perfectly fine with computers 10 years ago. Today I probably need to spend even more money just to continue to be fine at playing, working and living with them.

http://tinyurl.com/2j2cm9

At least for my wallet, computers are getting effectively more expensive.

Here in Finland electricity is ~0.05€/kWh. The market is deregulated, you can choose any electric company you want. The price includes an energy tax of 1 cent/kWh.

But "cheap energy for the industry" has been a mantra in the establishment for a long time, luckily consumers benefit too.

@totolamoto: I think the prices in France might be a result of the large percentage of nuclear power.

My home rig has a power supply with an integrated fan adjustor and wattage display that goes in a 5" bay. I've used on on my last 4 motherboards now. I find that I pretty much have to be doing something active for it to go over 100W. It almost never goes over 110, and when it does its typically because a game is starting up (3D card + Hard Drive + CDROM).

Typically it seems to idle in the upper 70's to low 80's.

My wife's machine, an old WinXP Athlon machine, is basically always on. I set it to standby/hibernate/whatever after 45 minutes of idle, but when it actually worked it was a pain because XP takes a looong time to go down and come back up. Recently something changed due to something that was installed (no clue what it is), and XP no longer powers down. There goes power saving.

I don't want to turn this into a flame war, but by contrast my mac goes in standby very quickly and comes back up instantly, and no installed software seems to interfere with the mac going to sleep.

Duncan Wilcox: "XP takes a looong time to go down and come back up".

This is generally caused by a driver that takes too long to answer to the stand-by request. My machine, with an Intel 865PE chipset and NVIDIA graphics, takes less than 2 seconds to go to stand-by or wake up.

Me: "This is generally caused by a driver that takes too long to answer to the stand-by request."

By the way, IIRC, Vista has improved this area. It'll wait for drivers for only a limited amount of time. Sleeping should be faster, then.

Wouldn't it make more sense for them to have one 120V to 12V transformer per server room, and then run 12V directly to each computer? Somehow it feels like you could do a better job that way, or am I missing out on something? Didn't someone at Google suggest that we should use 12V at our homes instead of 100/120/230/whatever volts?

There is more to this Google/power story that I don't fully understand. It appears that Google has been pressuring state legislatures to rewrite laws concerning power usage reporting so that they don't have to report how much power their data centers are using (see http://blogs.zdnet.com/micro-markets/?p=1277 for example). If they are making power usage so much more efficient, why would they need to hide how much they use? Speculation among some fellow geeks has been that Google corporate wants to appear "green", while sweeping under the rug the exorbitant amount of power they are starting to consume at their "out-of-the-way" datacenters.

This makes me think back to my Macintosh Plus. It cost $2,500 (in 1986 dollars), well above today's price for an entry level iMac. It consumed 60W and ran cool enough that it didn't need a fan. (Of course its CPU ran at 8 kHz.)

I run a laptop with a pretty serious CPU (Core Duo T2700 @ 2.33 Ghz) and a good, gaming quality video card (NVideo GeForce 7600). It is "rated" at 89 watts (4.7 amps at 19 volts) at the power supply, but I know that the computer it's self uses much less than that on a regular basis. The power supply is only slightly warm, and it can run the computer and charge the battery with the same 4.7 amps. I know that the computer can run without the battery installed, so that must be the peak wattage used. I run the laptop 24/7, though it usually snoozes (not standby, but video and hard drive shutdown) during the night.

I originally went with the laptop for noise reasons, but I do love it's low power operation.

At least we aren't all using those big 22 inch CRTs anymore. My old one was registered at 450 watts! That would warm up my small office in no time.

Power supply manufacturers could make their power supplies much more efficient and it would probably only add 10-15% to the cost. But if they are $1 more than the competition at NewEgg, they loose the sale. It's an uninformed market. If they put "85% efficient" in the title, it would sell more power supplies. But instead they put "455 watt" so they can win the "whip it out" contest.

Brainstorm: How about a little generator wheel in your mouse and piezo keys on your keyboard to generate power while you use your computer? Off to the patent process....

"how often is your PC operating at full load? If you're like most users, almost never."

I find this deeply, deeply sad. More people should be gamers! I am a gamer, and I run my computer at full load from the moment I plop in front of it, to the moment I finally stumble off to bed.

In science, computers are idle half the day. The rest they are actively running from photoshop filters to 3D rendering to massive scipy/Matlab matrix crunching or assembling images. Universities would benefit a lot as well from efficient power supplies, even if efficient only at the 250W+ range.

Google is brilliant, simply brilliant.
They have a power problem, that is very true. They also pay a premium for more efficient power supplies.
It is true that most home PC's are idle most of the time, so do not benefit from a more efficient power supply. But, if all PC's shipped with a 90% power supply, the supplies would be in volume manufacturing.
The premium Google pays would go away. They could buy standard (and efficient) power supplies off the shelf, for cheap.
Brilliant.
Plus, they look like good environmental Al Gore friends of the earth. PR and profits, brilliant.

Most of the world's computers are running screen savers at any given time (I'm not talking about server farms).

Some estimates put the number of mips consumed by screen savers at near 90%.

So, we are paying the power companies to run screen savers...

Is power so ridiculously expensive in California because some resource is running out? If so, there's a bigger problem than just money...

Does anyone know whether how many years we can power the USA using our current sources of fuel?

my cpu says up to 85% efficiency... but then i'm a gamer so i say waste away! my 8800 seems to suck a lot of power.

Glad to see Google pushing their innovations across the industry. The improvements they have to make by necessity can lead to a more efficient use of power by the entire population as better power supplies are pushed out.

With all of these issues people are beginning to see the true cost of inefficiency of devices on the planet. As someone commented above, one of the major issues is cleaning up devices that are still sucking out power on "standby" mode. If all the devices that are left idling could more intelligently drop their power use down then we would really see a substantial decrease in usage.

About two years ago, I was part of a group that was given a tour of DreamWorksSKG's computer animation studios, which included the "render farm" room full of rack upon rack upon rack of servers grinding away to render every frame of their latest films. The tour guide mentioned that the beancounters had been having a hard time settling on a way to budget the computer time against the films. The cost of the hardware bought for one film shouldn't be counted against a second film, and anyway the hardware costs were becoming a smaller and smaller fraction of the budgets. After some analysis (including the electricity/CPU costs), the beancounters settled on a metric that has been surprisingly consistent over the years, regardless of changes in CPU speeds or power consumption.

They now budget the costs of THE AIR CONDITIONING USED TO KEEP THE COMPUTER ROOM COOL. Frames of finished film rendered versus BTUs.

This was initially surprising to us on the tour, until we did as the tour guide suggested and held a hand up on either side of a typical server rack to feel the breeze flowing through it. Sure enough, the
temperature was quite noticably (maybe 15 degrees F) warmer coming out than going in. That's a lotta heat.

Makes me wonder if I should think of my computer as a space heater.

I've seen figures up to 30 cents/kw/hr for California. For those of you who don't think that's bad, I'm paying 6.8 cents in Virginia. That's the maximum rate for summer months. Of course, we get about 30% of our power from nuclear.

If I was in California I'd be rioting in the streets. I can't think of a reason for that large a disparity that doesn't involve corruption and/or incompetence.

Didn't IBM just announce something about power-saving datacentres?

A. Lloyd Flanagan stated: "If I was in California I'd be rioting in the streets. I can't think of a reason for that large a disparity that doesn't involve corruption and/or incompetence."

Apparently, you missed the power crisis in California a couple of years back. The problem is simple: NIMBY and BANANA ("Not In My Back Yard" and "Build Absolutely Nothing Anywhere Near Anything"). California basically failed to improve their electricity generating capability for several decades while power use climbed continually. At that time, the vast majority of their power came from natural gas fired plants. Natural gas is fine for handling peak loads. But, it's expensive. Using it to handle all the load is dumb. Coal, nuclear, or hydro is what they needed for that. They didn't have it, and AFAIK, still don't have it. Thus, expensive electricity.

> No, the real challenge is to drop power consumption levels on idle to something *much* lower.

Oh, it's certainly possible to build a low-power PC if you are selective about the parts you use. My home theater PC, for example, draws about 70w at idle.
http://www.codinghorror.com/blog/archives/000221.html

The other option, as Chubber pointed out, is to go with laptops which are low power by necessity. Since more and more people are choosing laptops over desktops-- within a few years, more than 50% of all computers sold will be laptops-- maybe this is a self-correcting problem.

I understand how you can use a meter (Kill-A-Watt or similar) to measure the energy draw from the wall socket, but how do you measure the DC output from the power supply?

A few years back I was participating in an effort to validate computer models used to predict climate change (climateprediction.net), to which I contributed 129.4 simulation years from 8 CPU's in my home that were otherwise powered on and running 24/7, but 99% idle. I stopped contributing when the 5% higher electricity bills started coming in--electricity bills in my area come with a handy historical consumption chart, so the real environmental effects of trying to predict climate change were pretty obvious. Oh, the irony...

Someone had a blog post recently that suggested that climateprediction.net and similar projects should in future use only Sony Playstations and video card GPU's, since the power-per-teraflop efficiency of desktop CPU's was so much lower.

My desktop at work has a CPU which in theory consumes 90W in "performance" mode (3.0 GHz) and 82W in "powersave" mode (2.8 GHz). The older machine it replaces had a 25W 375MHz mode. This is not progress...

I literally use my computers as space heaters. When working at home in the winter I simply close my office door to retain all the heat generated by the C++ compiler, instead of raising the thermostat to heat the entire house, and I have used my laptop instead of a blanket while watching TV on the couch on more than one occasion...

John wrote:

Wouldn't it make more sense for them to have one 120V to 12V transformer per server room, and then run 12V directly to each computer? Somehow it feels like you could do a better job that way, or am I missing out on something? Didn't someone at Google suggest that we should use 12V at our homes instead of 100/120/230/whatever volts?

No. There's this little problem of I-squared-R (Ohmic) losses. It takes more current to transfer a fixed ammount of power at a lower voltage (P= VxI). Unfortunately, wires are not superconductors, and their non-zero resistances end up heating the wires. The power dissapated in the wires is Pdiss= (I^2)R, and goes as the SQUARE of current. For a fixed power, doubling the voltage cuts the current in half, and reduces the power loss in the transmission lines by 75%!

Ever wonder why power transmission lines that cross the countryside are operated at such high voltages? It's to minimize the ohmic power losses.

> but how do you measure the DC output from the power supply?

Usually, you don't. You estimate DC output based on typical efficiency numbers for that PSU at that wattage level.

Places like http://www.silentpcreview.com are measuring DC output, I think using multimeter alligator clips (and a heaping dose of caution!).

A data center's raw power costs are not 15c/kWh. Every Joule of energy a server uses has to be removed. By AC this typically uses 60% as much energy again. Data center operators have to supply conditioned power. Accounting for the infrastructure cost for the UPS and secondary power source pushes the real cost of power way up. Consequently, the real cost of power in a data center is likely to be more than double the most expensive rate listed above. This halves payback time for any power savings anywhere.
I am currently helping to figure out how not to build another data center by making better use of our existing one! Good for business, good for the environment.

For a detailed breakdown of AC power demand under idle and max load conditions for many different types of systems, have a look at the updated REAL SYSTEM POWER REQUIREMENTS table on page 4 of SPCR's Power Supply Fundamentals -- http://www.silentpcreview.com/article28-page4.html

Most of the example systems are using high efficiency ATX PSUs; the highest one being 84~85% at the top of its efficiency curve. The max power draw of even an OC'd Pentium D950 + ATI X1950XTX system is just 256W. If a 70% efficient PSU was used for this system, the power draw would be more like 310W.

I used to use an old PC as my home server, but switched to a home-made low power box to save money. It cost me a bit under $500 to put together, but saves about $20 a month in electricity. The powersupply I use in it is close to 90% efficient... It uses a transformer to generate 12V dc and then uses DC-DC switching regulators to get other voltages. I measured power consumption at the wall and the whole thing uses 18W typically. When you consider that my old machine used 150W typically and still drew 7W when turned off, that's a huge improvement.

Yes, 14 cents per kWh is expensive.

Here in Ohio I pay about five.

14c/KWH is exspensive, here in Manitoba 95% of our power is Hydro power and we get it for <$0.06CAD

Maybe Google should move here.

http://www.hydro.mb.ca/regulatory_affairs/energy_rates/electricity/current_rates.shtml

Actually for large businesses it drops down to just above $0.02 per KWH if you own the transformer on site and use more than 100kv.

>For one thing, computer software starts to look incredibly expensive

The software that I've been using for the past 13 years is free to use (gnu, Linux, Debian, Ubuntu).

Bill's argument, what is you time worth, is lame.
My time is worth a lot -- the software is still free, I didn't pay to use it.

Jeff,

I liked your article. I'm surprised to see how much political fervor it has sparked. There are many different points of view about how much energy 'should' cost and whether market forces will self correct problems or whether people should artificially raise energy prices to encourage alternatives. Lots of it seems tied to the 'hot topic' of Global Warming and that political football.

No matter whether you're a died in the wool socialist or a capitalist, it makes sense to say that less is more in this case, people. Technology has always made gains by doing lots more with lots less.

g

batteries for my flashlite and headlamp are getting too expensive

10-15 years ago they were predicting that the US had coal reservoirs that would last over 100 years with increasing demand and w/o looking for new ones.

Stupid suggestion for all of those data centers:

1. Get rid of all but 12v there is no reason for the other voltages everything can be run off of it.
2. Get rid of power supplies entirely and have one AC/DC converter for the building that is ultra efficient and supply all of the computers DC.

#2 on the list is such a simple thing that should be done in all new homes, adding DC voltage. Everything and their uncle uses DC voltage in houses now except for refidgerators and stoves and even those can rn off of DC just as efficently if not more so than AC.

The key is that you need to use AC for transmission for effective transmission (just ask Edison). But in the house, there is no such requirement. We should be transitioning to DC from AC power in homes ASAP. The cost/energy savings would be huge... just think of all of the power bricks around your house, each and every one of them 70% effective, when you can create one that is 95% efficient (because it's so large) and get rid of all of that heat, and all of those power bricks...

About the 'anecdotal' evidence that Vista with Aero uses more power...

Tom's Hardware did a benchmark 1-2 months ago on this. They actually tested how long the battery lasted with Vista + Aero vs XP. It turned out it the two usedroughtly the same power.

While this might be surprising to some, it definitely is not strange. Video Cards are so powerful nowadays that the operations performed by Vista's Aero is something very simple and non-demanding for it.

You are missing the point that 12v is primarily used for mechanical components, whereas 5v and 3.3v are much better for logic. A change to all 12v would require:
Redesigning chips to work on more voltage
Making bigger chips to handle the extra voltage (wires or leads need to be bigger)
Requiring the proccesor to have its own voltage regulator, because all motors connected create interference.

And of course, ALL the chips in the system need to be redesigned. Not just the CPU. ALL of them. I don't particularly think that this is such a good idea.

Also, a PC operates on 12v, -12v, 5v, -5v, 3.3v and sometimes 1.5v (not provided by a regular ATX PSU, this is generated on the motherboard).

Why not just write better software that requires fewer servers all together? I know that MS and google are building giant data centers in several places and we assume this means that this will lead to a server count at google greater than the 400,000 number that was reported from a few years ago. This reminds of of the quest to build the heaviest airplane. Shouldnt we be looking a building a plane that flights farther, faster and on less fuel? I would really be impressed if Google or anyone else was telling us how many servers they unplugged or powered down.

I got a Playstation 3 a few months back, and was initially excited to join the Folding@Home project with it. Excited until I discovered that running Folding@Home (maxing out the CPU) was costing $30/month in power. (This is partly because it bumped me up one tier in PG&E's fee schedule...you pay one rate for "baseline", another for +100% baseline and even more for +200% baseline. This last rate was $0.33/Kwh on my last bill.)

The PS3 uses around 180 watts while at max CPU. Between it and my usually-on server (~140 watts) the den is often ten degrees hotter than the rest of the house.

Lately I've been wondering if there could be some way to capture all this wasted heat energy and use it for some useful purpose.

How efficient are laptop bricks are conversion? They too are power supplies.

Maybe time to kick the script and managed code habits?

Very interesting article... I'm curious as to what got you thinking about this?

Speaking as an electronic engineer, the suggestion to "simplify" power supplies to a single 12V power supply wouldn't actually make things any more efficient, in fact it'd probably make things worse.

When you want to change a DC voltage to another DC voltage, you've really got two choices :-

Linear power supplies: voltage output is less than voltage input, output so once you've factored in in-circuit losses the efficiency is a bit less than Vout/Vin (12V to 5V = 41% efficiency, 12V to 3.3 = 27.5% efficiency).

Switch mode power supplies: voltage output can be less or more than the voltage input, output current depends on the efficiency or the supply and the input current. Generally, unless specially designed these work better under higher loads. Efficiency depends on the load and design, but typically 70 to 80%.

Now it's just not possible to "redesign the chips" to live on 12V. This is how things worked in the past, but as the clock speeds went up and the transistor sizes reduced the supply voltages had to come down - not only was the voltage the major source of heating, but smaller transistors cannot cope with the high voltages. So nowadays, even though the power supply is outputting +3.3/+5/-5/+12/-12 you'll find that the motherboard is covered in small localised power supplies (usually a combination of switch mode and linear supplies) reducing the supplies to lower values like 1.2V/1.8V etc.

If you consider that a the average efficiency on each switch mode supply is 75 to 80%, it's pretty obvious that cascading just makes the losses increase. If anything, you'd be better off designing a single multiple output power supply that generated every signal you'd need - but this isn't possible given the way that the microprocessor cores, and core voltages change from generation to generation.

I'd have to agree that the best way to reduce power usage would be to run a simpler system: powerful CPUs, high speed HDDs, high end graphics cards come with high power requirements. Multi-core CPUs are an improvement, but I don't think it's possible to power down any unused cores (yet). Use LCD displays (typical LCD uses around 25 to 40W, CRT uses 100 to 200W, plasma TV uses 400W to 1000W).

On the idea of running a single 12V supply to power a building is just crazy - not only would you have problems with ohmic losses (as mentioned earlier) but you'd need to shift some really dangerous currents around the building - for a 1200W microwave oven you'd need to supply at least 100A (excluding losses). Not only is this impractical, but it's also highly dangerous.

Maybe that's a sign that Quebec will be seeing more datacenters popping up; Montreal's prices are like 6.6¢ per kilowatt-hour.

Apparently, what Bill Gates has said is very true. As the power supply increases over time, the demand for that amount of power has to increase. Otherwise, the efficiency (defined by power demand/powersupply x 100%) will drop.

Furthermore, the hardware prices have been dropping, despite the increase in complexity and sophistication of the design involved.

In my opinion, there is no clear cut relation between hardware prices and power supply, but somehow the graph as shown does depict a trend over time.

A number of companies besides Google are focused on developing high-efficiency power supplies for data centers, including ColdWatt (coldwatt.com) which claims to hit 91 percent.

everyones turning green. intel, google, shrek.

this made the front page of digg.

Use a script or hibernate, you can save at least 8 hours (1/3 a day) of energy your computer wasted. This saves more than a 90% efficient power supply. For google, i'll shutdown US server farm when the timezone goes to European.

[RegisterSaveEnergy.cmd]
set StartTime=22:00:00
set StartDate=
FOR /F %%d IN ('echo %date%') DO SET StartDate=%%d

ECHO "%WinDir%\system32\shutdown.exe" -s -f -t 90>"%Windir%\System32\Shutdown.cmd"
ECHO CMD /K >> "%WinDir%\System32\Shutdown.cmd"

SCHTASKS /DELETE IN SaveEnergy /F >nul 2>nul
SCHTASKS /Create /SC DAILY /TN SaveEnergy /ST %StartTime% /SD %StartDate% /TR "%WinDir%\System32\Shutdown.cmd" /RU "%USERDOMAIN%\%USERNAME%" /RP *

Performance/watt is indeed increasing, if you need proof of this then compare a 3.2Ghz Pentium D with a 3.2Ghz Core 2 Duo. Fully loaded that Pentium D will easily eat through 135 watts while the Core 2 Duo would be sitting somewhere around 100 watts tops. Now factor in the gaping performance gap between the architectures and the C2D is an obvious winner. With upcoming quad-core chips dropping considerably in price Google will see jumps in performance/watt consumed as you'd be able to load powersupplies at higher values, power fewer motherboards with denser rigs (1 quad-core vs 2 dual-core rigs) and cutdown on the costs of powering the excess peripherals. Processors are constantly surpassing their predecessors in both performance and power consumption.
ps: 3.2Ghz can usually be hit on the C2D with default vcore (1.25-1.35 depending on the chip)

@Tim
Exactly. My desktop computer has a 1.2v supply (most likely for logic) which is of course done by the motherboard, however, you are correct, redesigning ALL the chips for use with 12v would not work very well. I did say "Making bigger chips to handle the extra voltage (wires or lead need to be bigger)". 12v would burn out most chips today - even gas-guzzling TTL only takes 5v! Of course, CMOS WOULD take 12v, however, when was the last time you saw a CMOS ALU? Or accumulator? Or... anything beyond a few simple gates?

And of course, linear power supplies are huge! I have a 550w PSU in my desktop computer, it is the size of about 3 bricks. A 550w linear power supply would be bigger than the case!

And on the idea of 12v to power the entire building - forget about it! Why do you think we use AC now instead of DC? BECAUSE IT TRAVELS FARTHER!

Of what I said earlier about 12v running primarily the mechanical components - well that really is ALL that they do. The CPU is usually on 5v, more likely 3.3v or even 1.2v/1.8v (on mine, 1.2v). 12v would fry it. And that would be hard to work around.

Oh well... I guess you can't win them all...

Jeff,

Interesting analysis... but I disagree with your conclusion. Most people would think a 3-year payback is a fairly good investment. With the growing passion for green alternatives, I think people will be looking for conservation alternatives at all levels of their home and offices.

In “GREEN: THE NEW RED, WHITE AND BLUE,” Thomas Friedman, New York Times columnist and author of “The World is Flat,” visited the hydro-electric power plant that Google is using to power one of its data centers on the west coast. In the future, all technology companies need to address power consumption as part of their overall marketing and cost strategy.

Also, I have to agree with our European brethren that while 12 cents a kWh is high by current American standards, it’s just the beginning. I say this with a slight grimace as I current enjoy an artificially low 7-8 cents per kWh price here in Ohio.

For servers, HPC, and datacenters - it's all about SWaP.

Those same metrics can also be applied to home power users where (in my case), I'm running 8 systems 100% load 24/7/365 of a 20A breaker.

Another note that people often don't realize: Intel processors (STILL) consume more power than AMD.

Final note (that also people don't realize):
The cooler your CPU/system runs, the more efficient it gets. Less current leakage. (Based on data/results/publication from the IBM Journal of Research and Development for the z9 system).

Gregory asks: "Is power so ridiculously expensive in California because some resource is running out? If so, there's a bigger problem than just money..."

The resource that is running out is generating capacity. No new power plants have been built in California since the 1970's. We have actually lost generating capacity since 1982 as older plants go offline (either for more frequent repairs, or permanent shutdown). California's population has doubled since 1970. California's electricity consumption per capita doubled between 1980 and 2000. Twice as many people using twice as much electricity per person yields a fourfold increase in demand.

No new generating plants were built because of environmental impact concerns, long regulatory delays, and community activism. California has been forced to purchase electricity from out-of-state generators for years, often at unfavorable rates - so unfavorable, that it doubled or tripled our electric bills in a single year, costing the governor his job. ["Unfavorable" is a euphemism - Enron was charged with criminal price gouging in its sale of power to CA before its collapse. They were ordered to refund a chunk of the take, but went out of business before paying up.] After the California electricity crisis peaked 5 years ago, the governor has fast-tracked the construction of new generating plants, but they are not expected to go on-line for several more years.

In the long run, the falling cost of computing hardware is largely driven by Moore's law, which has lowered watts per MIPS at roughly the same rate as it's lowered dollars per MIPS.

It's true that notebooks use a lot less power --- typically 10-40W rather than the 200-400W figures you're bandying about above.

A three-year payback is nothing to sneeze at. That's a very-low-risk 33% ROI. You aren't going to find that in the stock market!

Remember that every watt you use in a computer in a data center becomes 2-4 watts when you include the power draw from the air conditioning. This is also true in your house when you have the air conditioner on, but if you are paying to heat your house, it just means you have to run the heater less. Whether this saves or wastes money depends on what the heater runs on --- if it's electric, it's obviously a wash, since 200W of heating is 200W of heating regardless of whether it's operating by heating a CPU or a nichrome wire. Most other heating fuels used to be cheaper, but I think the recent rise in the price of natural gas in the US has made it more expensive than electricity.

RC, Google wants to hide their power usage because they don't want their competitors to know how much computing power they have. It makes it more difficult to enter a market and compete with Google if you think you can do it with one-tenth the compute power they use (especially if you're wrong), or if you think you need ten times the compute power they do and therefore overbuy.

To those who asked why power is expensive in California (50% more than in Finland or most other US states), the answer is that the utilities there were forced to sign long-term power contracts in 2001 in order to stop getting hammered by rolling blackouts. These contracts require them to pay extortionate rates for power. It turns out that the power shortages were engineered (by Enron, among others) precisely to extract more money from California. After these crimes were uncovered, if I remember correctly, California's governor Gray Davis initiated a lawsuit against the energy companies responsible; they responded by funding a recall campaign against him and an electoral campaign by their buddy Arnold Schwarzenegger, who dropped the lawsuit as soon as he became governor.

California was particularly vulnerable to such machinations because (1) they are heavy energy users (because they are a major part of the national economy); (2) they don't have nearly enough local power production; (3) they had just voted against George W. Bush.

So, Flanagan's comment about how such high power rates must come from corruption or incompetence --- they were right on target.

This is part of why I'm in Argentina.

To those of you saying google should just shut off servers when not in use consider this. googles best use of their power is not in responding to queries but in crawling the web for more or updated content (a never ending mission)

To those suggesting one power supply per building you must by now realize the loss of power ofer the distance would be terible, however the use of one per few motherboards would not be a bad idea.

Google could save a significant amount by giving their programmers little terminal boxes (100Mhz and 32mb ram no HD or CD)and giving them access to a server cluster for their actuall use. This would incease the efficiency of each users individual machine by eliminating 95% of it's power. This theory would work for us at home as well, as connection costs go down we can start centralizing computing power and make more efficient use per watt. (now if someone would just offer this as a service....)

I think the real key is to have more of this power supply I read about recently that only runs at whatever you need. So you can buy a 400 W power supply and it will run at 150 W idle, 200 W with some stuff running, and 400 W with the graphics card, 4 DVD burners, and 2 HDs or whatever - I'm just making up numbers, but the concept is great.

ERM -- that's how all power supplies work. They only use whatever power your machine is currently drawing.

For the record, computer hardware is not free. Kind of like those hardware reviews in the magazines and such (in which all hardware is wonderful), he's just saying "It's all good." Pretty meaningful, huh?

I am fairly certain what Bill meant is that the amount of power the end user is getting for their dollar almost makes it seem free. But you are right... computer hardware is not free... nothing is... ever.

Jeff Atwood wrote on May 23, 2007 09:40 PM ---

For the record, it is possible to design a power supply that's efficient at low load levels. But it's far from common.

http://www.silentpcreview.com/article263-page4.html

The Fortron Zen is pretty amazing; 77% efficiency at 52w, going all the way up to 88% efficiency at 300w.

-----------

Here's another PSU that does even better. Seasonic SS-300SFD 80 Plus -- 81.5% at 41W load; 85.2% at 203.7W load. Overall, the most efficient PSU we've tested.
http://www.silentpcreview.com/article286-page4.html

You state that "cheap hardware is a strategic advantage" for google. Don´t kick me for being an economist, but this is not an advantage for google, because it is something everyone can have. It therefore is not an advantage for Google.

Over the past year or so, I've become more aware...or rather I have kept energy usage more in mind than usual. I'm thinking Green...YES, YES I know, not another one of those. I'm finally looking into switching over to Linux for two reasons: The competition will be good (eventually) and lighter OS's (when you get to know Linux and tweak it). So, you (Jeff Antwood) calculated and mentioned that switching to more efficient PSU's is not yet a personal thing. I think it is every bit every single computer users priority. Being an Industrial Engineer, we know that from the Toyota Way of thinking, every single small step to better a system, is a good step. Yes, maybe the very efficient PSU's are more expensive and we wont see our returns in terms of electricity usage, but we could calculate the savings in the bigger picture: 15% more efficient PSU in your PC x (how many pc users world wide?). And to go back to my soap box, Global Warming might be a huge lie from oil companies to get into the energy business (non-fossil fuels), but wont it be nice to get up in New York and have clear skies? Thing is, as you stated yourself with your single pc cost saving, Global Warming will continue because it does not touch us personally.

Anyway, great post, found it very interesting and informative. All the best.

South Africa

Bought a kill-a-watt power meter, found out the my box is using 5.1kwh per day. That's $30 buck a month running 24/7. I will be shutting it off at night now.

eXtreme power supply calculator is what I use to get an approximate wattage for my new power supply

http://extreme.outervision.com/psucalculatorlite.jsp