Companies Waste $2.8 Billion Per Year Powering Unused PCs

snydeq writes "Unused PCs — computers that are powered on but not in use — are expected to emit approximately 20 million tons of CO2 this year, roughly equivalent to the impact of 4 million cars, according to report by 1E and the Alliance to Save Energy. All told, US organizations will waste $2.8 billion to power 108 million unused machines this year. The notion that power used turning on PCs negates any benefits of turning them off has been discussed recently as one of five PC power myths. By turning off unused machines and practicing proper PC power management, companies stand to save more than $36 per desktop PC per year."

Re:PC power management sucks...

[Sycraft-fu]

Something is wrong with your PC or its setup. There is no reason it should draw more than a small trickle when shut down. Mine measures 0 watts when shut down. Now it isn't actually zero, the PC does draw a tiny bit unless I throw the hard switch on the powersupply, but that means it is less than a watt. That's how it should work when actually shut down. There is only a tiny bit of power drawn for things like charging the battery and the ability to do wake-on-LAN and such. 19 watts sounds like you have it suspended or something. Where it has shut down a large part of its components, but is still running in a low power state (RAM is being refreshed and such).

So this isn't a PC power management problem, this is a problem with your particular PC.

Re:Turning PCs into a grid

[terraformer]

You are absolutely wrong. A system like that is the exact opposite of efficient if you are leaving the PCs on *solely for the purposes of the cluster*. Grid computing is only efficient when you take what is under utilized and put it to work. The energy in any PC (server or desktop) can be split into two parts. Overhead and active. All of the overhead is what the PC consumes when it is idle (~0% utilization). All of the active is the power between idle and max consumption. On your typical desktop PC, more than 60-70% of the energy is wasted overhead (conversion losses and platform power). By amortizing that overhead between TWO tasks, and putting the CPU/GPU to work when sitting there idle, you are far more efficient than when simply at idle. ie; You are still benefitting from the idle by having a ready PC to compute your next command AND the work performed for the cluster. But you are not more efficient per work unit than a super computer with it's low relative overhead when only running ONE task, the cluster computing. This is because super computers have more overhead but they can amortize that overhead over a large number of CPUs with a high compute capacity.

Re:BUY software to shut down a PC??

[totally bogus dude]

Not so good with Vista though, as the warning dialog appears in another desktop. Part of that secure desktop thingy for UAC prompts and the like. You get a program appear in the taskbar but unless you actually notice it and click on it you'll never know your PC is about to shut down.

Your basic point is correct though, but I think a lot of organisations prefer to buy stuff than have in-house staff capable of writing even simple scripts like this. Presumably it's for the same reason they'd rather pay some consulting company loads of money to build an SOE we could've done ourselves: if it's outsourced to a high-priced company, it must be better!

I didn't RTFA, but does the product they're suggesting produce pie charts? That's probably the answer.

Re:Half an hour a year?

[daem0n1x]

That's why I use hibernation. It takes one minute to shutdown and another to startup.

Re:Turning PCs into a grid

[jw3]

The computers are not being left *solely* for the purposes of the cluster. The policy of the university admins is to leave them overnight for updates, and anyway the users don't like to turn them off (so they don't have to wait for the computer to boot up in the morning). Therefore we are utilising what sits there idle anyway. Furthermore, anyway you don't take into account the overhead of buying a supercomputer / cluster with 1000+ nodes in the first place -- and we are utilising what has already been payed for (both in terms of money from the university and in terms of energy used / CO2 emission that took to produce the units). Finally, buying a supercomputer / cluster is, due to the necessary bureaucracy involved in expensive investments, a major pain in the ass and also a system-administrative effort.

Of course, this solution cannot replace a proper cluster -- I have already outlined why, and also I agree with you in puncto efficiency. But if you have a bunch of PCs sitting around idle at night, and need calculations -- this may be a cheap and quick solution.

j.

Re:Turning PCs into a grid

[walshy007]

1. paragraph tags make posts as long as yours easier to read, for future note

2.Your essential point is it's more efficient to use one presumably NUMA supercomputer to complete a task, which may or may not be the case depending on the supercomputer and the task given, but the point is.. they don't have a supercomputer, and likely don't have the funding for one.

Using their spare pc's at night in a clustered environment would be one of the most cost-efficient things they could do in so far as hardware purchasing, considering they already need and have the pc's setup in the right configuration

we don't all have a 128-cpu onyx 3800 gargantuan tower sitting in our closets for this kind of computing, we do tend to have at least a few relatively fast desktops available which would otherwise be off or idling.

Re:I didn't really get this at first.

[Migraineman]

I disagree with that last statement. I'm doing some embedded development right now, and I leave my machines on overnight intentionally. Powering up the entire system to the point where I can continue from CoB yesterday would take 30 minutes or so. I'd chew through that $36-per-year savings in a few days, possibly one day if I'm working at a customer's site running at my external-billing rate rather than my internal rate.

And no, I can't just go get coffee while the machine boots itself. The applications interact with the target, and get completely hosed if the host or target machines go into power-save modes.

Re:Magic smoke

[hcdejong]

28.8 billion kWh/year is more than enough to 'change the power plants operating conditions'. A 125 MWe unit (the output of one generator of a nearby power station) delivers about 1 billion kWh/y, so shutting down all PCs at night would make a significant dent in the base load.

Heat Cycle Bullshit...

[cyclocommuter]

I assembled an AMD Athlon / Athlon ASUS A7N8X and a Pentium 4 / MSI motherboard powered PCs at about the same time more than 5 years ago and these computers are being powered on and off almost everyday. They still work.

Newer PC components especially the motherboard usually still have juice in them even though you power them off. The CPUs and graphics card even when powered on will still experience heat cycles ranging from just above room temp when idle and depending on the efficacy of the cooling system, to 60 C (for CPUs) or 90 C (for high end graphics cards) when playing games.

Re:Magic smoke

[necro81]

Electricity is a fluid commodity. Although long-distance transmission losses mean that power is consumed mostly near to where it is generated, I wouldn't make a blanket statement. The study is, after all, making generalizations about computers all over, which are powered by a mix of energy sources. So, the emissions attributable to them should take that mix into account.

In making these kinds of calculations, I'd just figure the energy makeup of the entire United States [wikipedia.org] (or whichever country you prefer). For the U.S., nuclear makes up about 20%, natural gas another 20, coal about 50%, hydro about 7%, and other renewables about 2%. So, in figuring the carbon emissions for electrically-powered equipment, I'd say that for every Watt-hour, 70% of it produces carbon dioxide directly, and the other 30% can be discounted.

In other words, unless you want to talk about very specific cases (e.g. the off-grid guy who powers his home and computer from photovoltaics), no one is completely clean; everyone is about 70% dirty.

Besides, even if you want to declare your virtue by powering your idle computer from nuclear energy or magic fairy dust, do you really want to boast how clean your wasted power is?

Re:Magic smoke

[NeverVotedBush]

Why "5 minutes"? I would guess that if you turn off PCs after the workday and don't turn them back on until the next morning you save more like 15 to 16 hours of run time.

That's 960 minutes per day x 230 work days = 220,800 minutes. Or 3,680 hours per desktop per year. That's not counting in the 48 hours every weekend (52) which equals an additional 2496 hours, plus however many holiday days at another 24 hours each. If there are seven for whatever business, that's another 168 hours. And if the worker takes off two weeks each year, that's an additional 336 hours.

Grand total is 6,680 hours of wasted run time as an estimate.

For the people who run the fancy screensavers, the power used is fairly large. A blank screen is the best. That lets the monitor go into low power.