> i'd bet that there are enough ppl with these setups (i still recall the guy
> with 8 physical cpus in one of his boxes)

it would be fair to assume though that having a farm setup like this would require less power consumption than 8 seperate boxes.

---

Kolch - Crunching for the BOINC@Australia team since July 2004.
Search for your own intelligence...

well, in my case I am unfortunately and very sadly single (and seemingly forever), therefore my bed doesn't get used for that other use. Add this to the fact that I work 24 hour shifts and that cuts out another big chunk of time my bed doesn't get used even for sleeping. (except on my days off)

Jim

> Nah nah nah...theres a problem with all of this.
>
> With all of your concern about whether the bed was radiating enough heat, and
> trying to keep the bed cool, you would avoid other uses for the bed itself =P
>
> I've heard of video games creating this problem but never BOINC...
>
> Hmmm...maybe its an alien plot!!
>

---

> 17 'puters.... 40 watts each....


.68 kilowatts, that's about .007 cents an hour and I live in the swamps of Louisiana.

.

---

<a href="http://www.brainsmashr.com"><img src="http://www.brainsmashr.com/signature.gif"><img src="http://brainsmashr.com/boinc/counter_big.php?id=305369&amp;project=seti&amp;ctx=white&amp;cva=red&amp;cbo=white&amp;cbg=black&amp;linethickness=2"></a>

> One way to cut down power usage if running seti (or most other BOINC enabled
> projects) 24/7 is to run it in a ramdrive, have all your harddrives spindown
> when they are not used and have enough RAM to avoid swapping.

Even without a RAM drive: My PC has 4 harddisks in it, so only one needs to be spinning for BOINC (and my satelite receiver actually also uses it for storage, so I want to keep it running so that timer recordings don't fail to start because the hdd wasn't running). Would there be a way (in windows xp) to set the power management for the other three drives to spin down, but keep the one running? So far I've only been able to either have them all run, or have them all spin down...

---

In light of the current conversation between Administrator and others, I've dredged this thread up from 452 days ago.

Earlier in this thread stated:

Message 63208 - Posted 12 Jan 2005 3:32:14 UTC
Last modified: 12 Jan 2005 3:38:35 UTC
OK, I went and did it. I placed my multimeter in series with the cord and as expected the "Clamp On" ammeter wasn't very accurate at that low of a current draw.

This puter drew from 830-912 mA (milliamps) from Bios load to the completion of windows loading. I shut down Boinc and it dropped to between 630-655 mA. This represents a delta (difference) of 228.5 mA (avg)or 27.83 watts.

I'll start boinc. Yep it again went back up to 884mA. It is consistent.

So, running this puter with Boinc ON costs (871mA [avg] x 121.8 Vac= 106 watts) 106 watts/1000 x .08cents=.00848 dollars/hour or .848 cents/hour. .848 x 24 = 20.352 cents/day or $6.11 a month.

I could save .22264 cents/hour, 5.34 cents/day, or $1.60 /month. heck, If this puter wasn't crunching Boinc work I could just shut it off and save the big $6.10/month.

Now these are accurate figures (unless I made a math mistake) and can be used for other purposes.

Now for my own curiosity, I've done the same to all my puters and here's what I found out:

AMD 64 3700 Sandiego, Asus A8N-E mobo, Enermax 450W psu, Asus 6200 PCI express video card, 1M OCZ gold ram, OCed.

With boinc running 1120mA max, 1024 mA min, 1043 mA average (125.16 Watts)
W/O boinc running 888mA max, 704 mA min, 704 mA aversage (84.48 Watts)
This machine costs 1.25 Cents/hour, 30.04 Cents/day, and $9.01/month.

Gateway laptop, AMD 64 3700 754 socket, 1 M ram

With boinc running 1200mA max, 752 mA min, 1157mA avgerage (138.84 Watts)
W/O boinc running 496mA max, 416mA min, 440mA average (52.8 Watts)
This machine costs 1.39 Cents/hour, 33.32 Cents/day, and $9.9965/month.

HP 760n, P4 1.8, 512M ram, crappy Asus board

With boinc running 1680mA max, 1472 mA min, 1597 mA average (191.64 Watts)
W/O boinc running 1016 mA max, 928mA min, 949mA average (113.88 Watts)
This machine costs 1.92 Cents/hour, 45.99 Cents/day, and $13.798/month.

HP 6545c, Celeron 500, 256M ram

With boinc running 688mA max, 640mA min, 672mA average (80.64 Watts)
W/O boinc running 648mA max, 448mA min, 460mA average (55.2 Watts)
This machine costs .8064 Cents/hour, 19.35 Cents/day, and $5.806/month.

conditions
All measurements taken with Fluke 87 multimeter place in series with the power cord to the puter.
All monitors powered seperately.
Measurement sample period 2-3 minutes of record time after allowing some "settle" time after turning off boinc, or after starting boinc.
Each sample repeated and all results similar to those listed, so I went with the higher Max sample on each reported result.
Monthly rate calculated with 30 days/month.
All rates calculated at 10 cents/KWh.

---

Now for my own curiosity, I've done the same to all my puters and here's what I found out:
****

Thanks for the info.
Interesting to see that the laptop with a very similar CPU to the desktop actually uses slightly more power than the desktop, and that the P4 really does suck up a lot of power, even when running at low clock speeds.

All rates calculated at 10 cents/KWh.

Oh to have such cheap power. Around the 14¢/kWh mark here.



EDIT- fixed multiple spelling errors.

---

Grant
Darwin NT

OK OK I was guessing. I pulled a power bill from my co-op and I'm being billed 12.21 cents/KWH plus a 1.5 cents/KWh "power purchase correction factor" (whatever that is). so it's really 13.71 cents/KWH

So the costs, should be 37 percent more than I've depicted.

OR

AMD 64 3700, $12.34/mo
Laptop AMD64 3700 $13.69/mo
P4 1.8 $18.90
Celeron 500 $7.95

NOTE: this bill was from August last year, rates have probably changed, but I'm not going through this again any time soon.

[edit] that's $52.88 a month or $634.56 a year donated to boinc projects.

that's $52.88 a month or $634.56 a year donated to boinc projects.

Interesting.

I have a challege for the crunching community.

I know it was half-assedly done below, a year ago, but, can we figure out how much power (kWh in a day, month, or year) the BOINC community uses? (or maybe just SETI, for now).

Proposals on how to do this anyone?

Last year (down in this thread) it was done by assuming an average computer, number of users, running 24/7. Which is innacurate on 3 variables. 1) "average" computer power usage 2) Number of hours the average person actually keeps their computer running, 3) How many computers the average person has running.

The latter two are probably the hardest to estimate.

Can we do better? Does anyone have any ideas? I'll kick things off. Here's my proposal, try to refine it:

1) Figure out how many work units on average, done per day, recently.
2) Figure out the average power consumption to calculate a work unit.

Then multiply the two together.

We'll still need to somehow arrange to choose an "Average Computer". I think that'll be our biggest variable.

But, this way eliminates the number of hours people run, a user count, and a number of computers each user has... because it's based on workunit. Workunits we *can* get an accurate number on, and that number incorporates all the other variables having to do with users and time run.

Anyone have any idea how to calculate an "average computer"? More specifically, we're looking at how much power it costs to crank out a workunit on average.

mmciastro figured out in the other thread that there is a non-linear relationship between power used and processing power. For example, his new AMD64 3700 box crunches 10 times as many work units as his P3-500, but as you can see below, has not even double the power consumption. So he roughly calculated it to being about 5 times as power-cheap to crunch a workunit on the new vs. the old machine.

So, adding to the difficulty of trying to pick an "average computer", if you're off by a small amount, it'll be a non-linear amount and you can easily end up waaay off by a large amount because your errors will be magnified.

That said... I think we can actually calculate that amount. We can at least ballpark it and establish a relationship. There is a statistic for "CPU-time used per workunit", am I right? That means we know how long a computer was on, to crunch a workunit.

But that *also* tells us how fast that computer is.. and it wouldn't be too hard to plot power-consumption for various types of computers. I.E. A P3-500 draws 80W, where an AMD64 3700 draws 125W. So we can say "This computer took 6 CPU hours to crunch a workunit. That means, it's probably a P3-700 according to our charts. That means, it probably uses about 100W according to our charts."

It's probably more accurate to plot it on an individual scale like that, then amalgamate the results at the end. But, if we have access to the user database or if this is in a massive datafile somewhere, another way would be to plot CPU-hours/workunit on a graph, and then use that graph to create an equation. I'm not familiar with what that would entail and the calculus that it would likely involve, so I'd go with the whole charts/lookup kind of thing. But, lots of smart people here who can tell us which is better and if they can do it the hard way.

There are lots of ways we could further refine this number by adding more variables. Most of these variables would be pairs of ratios.

Ex) AMD processors are on average 30% more power-efficient their equivalent Intel chips, and on SETI 40% of people use AMD chips. Lets first assume all chips are Intel, then modify. So those 40% being 30% faster (0.40 * 0.30) means SETI is actually 12% less costly than previously estimated.

Ex2) How many people are running optimized apps? What kind? How much do optimizd apps affect crunch time? Be sure to include this number in our calculations because it would affect how we've classified that computer for power consumption (which is on a non-linear scale!).

Tonnes of things you can do like that.

Another interesting number to figure out, would be, what is the *ADDED* cost of running BOINC?

Ex) I leave 1 computer running 24/7. I leave it up to pick up chats, transfers, and because I'm too lazy to wait for it to bootp every time I want to use the computer. So in this case, the *added* cost of running BOINC is much smaller than the number we had before, somewhere around only 40W. But, I have another computer running I would never plug in at all. So, it's *added* cost of running BOINC is the same as total cost.

Or, maybe I have a computer that would be on all day during the day anyway, but off for the night. So the added cost of that computer is 40W * 16h + 120W * 8h. Because I leave it on for the night.

So, if anyone can ballpark what percentage of computers (or to further refine, what percentage of what kind of computers, as the older ones on people's accounts are less likely to serve a functional purpose, and older computers are less power-efficient), and percentages of time are *added* cost.. that would actually be a far more useful number.

--
Okay, so at the end, we have these two numbers. Some amount of kWh per workunit, and the daily kWh that SETI consumes.

We can then try to figure out what the average cost of power is.

Any one of us could look on our power bill for all the variable costs, but I'm sure there's stats we could look up. This number could be as refined as we'd like it to be. We could just use "Average for first world countries." Or we could break it down to USA vs. Germany, multiply average US power cost by number US members, add to German power cost multiplied by number of German members.

We could do any number of countries, not just 1 or 2. Or, maybe USA and Germany have equal numbers, but Germans are running slower machines on average, meaning they crunch less workunits and use more power per workunit. So maybe each user should be evaluated per country. It would depend on fluxuations in Euros vs. USD if we wanted to go that far.

We could figure "Well, power cost actually drops at night in some places, so, 8 hours where it's cheap, how many machines are running then, that have that, in those places and what would our savings be?"

*Lots* of ways to make it more precise.

But... lets not to scare anyone off. We can start with some simple number someone can probably find in 5 minutes of Googling.

So we'd have how many kWh per day, and the average cost per kWh. Multiply the two together, and we'd know how much SETI costs per day in power to run.

That'd be neat to know.

---

Take it one step further. Does anyone have any familiarity with supercomputers? BOINC projects use the community of little machines to do the work of a supercomputer, right?

I am curious how efficiently they do so. Compare the cost per workunit in the BOINC program, vs. the cost of owning a supercomputer to do it instead.

I assume we're less efficient than a supercomputer, but maybe not, especially if we estimated the *added* cost, not just the raw cost.

That would be an intersting thing to know.

Now, here's where I get a bit crazy...

What about the cost of *renting* a supercomputer? People who know about supercomputers, does such a thing occur? Do companies buy time on supercomputers, or are they mostly just set up and wholly-used by the company that owns them?

The price of renting a supercomputer is a good comparison, because it tells us the actual value of the work SETI does. It compares us to the alternative.

Supercomputers need fancy buildings, maintennance staff, power of their own, and all sorts of things like that. SETI does not, it only needs a few people at the top. But supercomputers are designed to be run as such, and run much more smoothly?

Personal computers will fail or have components die after a decade perhaps (or maybe 3 or 4 years?).. so are cheaper because they're built to be disposable. Supercomputers, like servers, are generally not, so they're more expensive.

Running SETI on personal computers has some impact, but not much I suspect. Most people get rid of their computers before their CPU, HDD, RAM, etc actually out and die on them. Plus, it gets cheaper by age. If SETI kils your computer after 6 months, you'd be mad, that computer still has high value. If running SETI causes your computer to die after 3 years, instead of 4, not so bad. What is the price difference between a 3 year old computer and a 4 year old? Not much. And that's only on the percentage of people that it'll actually break on.

....

The reason comparing to supercomputers is interesting to me, is for SETI fund-raising concerns.

What if...

SETI whored out? What if it said "Some percentage of the time, the workunits you get will not be SETI, but part of a science app that SETI has sold computing time on"?

Sure, any company could start up a BOINC project themselves, but they'd have to recruit users. What if they didn't want to bother with that process, and just wantd to pay "A Supercomputer" to do some calculations for them? They could go rent time on a supercomputer, or, they could rent time SETI.

Does anyone have an idea if there is a market for this, or what the "renting" market for supercomputers is, if such a thing occurs?

Here's what I'm thinking...

SETI needs 2 things to work well. 1) Crunchers. 2) Money.

SETI has trouble getting money. To the point of "We may have to shut down the project!"

What if #1 took care of both needs?

....

Lots of people will get twisted panties over this, so, let me say it instead of you, you could opt in. Not mandetory.

How about that?

Now.. here's the kicker. SETI could compete with supercomputers. Sure. Because SETI has no actual crunching cost. It gets its crunching for nearly-free (just servers). So yes, SETI as a project *could* sell crunching time for cheaper than supercomputers.

But, how efficient would that be for the community?

The community donates their power bill. That's their mandetory donation to be part of SETI. Since people don't see "Cost of SETI" on their power bill, SETI gets away with a lot I suspect. It being a hidden cost is why so many do it. If at the end of the year everyone in the project was instead approached by a canvasser saying "Can you donate this much to SETI", where the amount was their power bill of running SETI for the year, most people I think would slam the door in their faces.

My concern... what is the value of supercomputing, and is it even less than the added cost of SETI?

What I mean is, say we have that number, the "Power cost of SETI in one day". What if we took the number of workunits crunched that day, and paid for a supercomputer to do it instead? Which number is lower?

Yes a supercomputer has its own power bill and all the rest of that, but it will also be more efficient. How much more?

What would be great.. is if people could donate $10 of power, and gain SETI $13 in revenue. That would be if the value of crunching on a PC supercedes it's power cost.

But what if people donated $10 of power and SETI could only sell the crunch time there for $7?

In my opinion... SETI should still go for it, because it needs the money. Even though it's more efficient to have directly given the money to SETI instead. Even though it's a net loss and burning $3. It's a bit of slap in the face to the community, but, it *is* seamless and they could still get away with it.

More rambling...

Is there a market for this? What do people want? It could be set up as a ratio of workunits, but, maybe clients would think that was too slow. So they'd want SETI's crunching power, 100% for a month perhaps. Then SETI would go back to its own, 100%.

Maybe SETI could spend the first 3 months of the year whoring out, to pay for the other 9 months of the year? Pay the bills first, do the crunching later.

Lots of options.

Or how about this... assuming the value of computer is higher than the power cost for your computers... what if an entreprenuer stats a BOINC project. They are a BOINC-based crunching reseller. Anyone who joins up crunches for whatever projects that reseller can sign on. The reseller takes 10%, the users take 90%.

So, you have an account with them, and for howevermuch credit you have, you get a cut of the profits. They paypal you or write you a cheque once a month for your credit.

Screw *donating* your idle CPU times, what about *selling* them?

What if you made $20 for every $10 of power you used? That means you could set the reseller project to 50%, and all your other volunteer BOINC projects to split up the other 50%, and then BOINC would pay for its own power bill!

Older computers would have to pick a high ratio, and some computers I suppose would not even be able to pay for themselves. But newer computers? Sure.

Ta da! Now how many people sign up?

Now, BOINC is open. Are there conditions for making money using it's code? Even if not, some of you are going to hate anyone making money off this, so how about this. What if BOINC set a monopoly and had the one and only reseller allowed being them?

And what if they picked a ratio, to prohibit anyone from turning a profit? So, if you can make $20 with $10 of profit, you *must* donate 50% of your resources to volunteering.

Actually.. that's probably just 50% worse than what they have right now. I don't think many people have quit or not wanted to join BOINC because of their power bill, because it's transparent. If you want a big signup bonus, maybe let people make *some* money on top of their power bill, but donate some time too.

Only problem is the power bill will depend on computer. So, maybe SETI would have to do the same thing we're proposing above, to plot probably what their computer uses for power, and be more generous to the people running older machines.

------

Anyway, the project and challenge being, can we figure out the power cost? All the rest of this just gives us information, not saying it should or will be done, merely to suggest whether it *could*.

What possible stats could we have access to that would help us calculate added power cost for BOINC? Can anyone tweak my proposal or suggest how to implement it?

Anyone interested in picking up the torch on this?

Administrator, why do you want someone else to do your research? You seem to know what info you are looking for so get off your duff and do it.

---

98SE XP2500+ @ 2.1 GHz Boinc v5.8.8

And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer.

See what happens when one answers one question???? It spawns 1/2 a dozen new questions. Almost makes finding the answer a bigger pain than a relief.

In my case if I weren't running Boinc this machine would be off about 20 hours per day. It's off only one or two hours when I'm out walking, running, running an errand or visiting my parents. Besides afternoon activities I have it play Internet fifties music in the midmorning and while preparing supper.

---

Anyway, the project and challenge being, can we figure out the power cost?

mmciastro already has if you check his post before my previous one.

---

Grant
Darwin NT

See what happens when one answers one question???? It spawns 1/2 a dozen new questions. Almost makes finding the answer a bigger pain than a relief.

Necromancing can do that to a person. ;)

---

me@rescam.org

In my case if I weren't running Boinc this machine would be off about 20 hours per day.

My hosts (my employer's computers) are on all the time anyway. In general I prefer to leave systems running, as I believe that starting up & shutting down causes unnecessary 'wear & tear', mainly from repeated cycles of warming up & cooling down. Moreover the Mac systems run various housekeeping tasks overnight. The monitors switch off automatically when not in use; CRTs can use more power than the computer itself.

---

Administrator, why do you want someone else to do your research? You seem to know what info you are looking for so get off your duff and do it.

"My" research? How is this "mine"?

I know what my personal cost per workunit is. That's trivial.

I proposed a community project to figure out what the cost and the value of the entire SETI community is. Information that might be used, within the structure of BOINC, to solve SETI's funding issues.

That's not "my" research for my benefit.

If there's no interest in what I suggested, that's okay, we don't do it. I certainly won't be handling the task myself. I have no idea how to pull all those stats or write an algorithm to analyse them, nor do I own several dozen machines to benchmark power consumption per workunit.

I suggested the project and did as much as I could to develop a methodology to work towards sovling it. What I suggested can be improved upon massively, by people more knowledged and more familiar with SETI. Hopefully there'll be a lot of feedback. If not, that's okay, no project.

... and to the people mentioning how often they would keep their machines running, this isn't the time. Not yet. First we need to refine our strategy before we go out and collect data.

In my case if I weren't running Boinc this machine would be off about 20 hours per day. It's off only one or two hours when I'm out walking, running, running an errand or visiting my parents. Besides afternoon activities I have it play Internet fifties music in the midmorning and while preparing supper.

In my case, if I weren't running BOINC my machine would be on about 16 hours/day.

Instead, because of BOINC, it's on 16 hours a day.

---

Are we forgetting the cost of an Internet connection, or at least a share of the cost as only Setiathome needs access 24/7. And then there is the cost of air conditioning, almost year round for me, to remove the heat generated by Setiathome computers. I figured out that my star would only cost me about half a month's expense of running setiathome for them free of charge, so why not give some directly.

Are we forgetting the cost of an Internet connection, or at least a share of the cost as only Setiathome needs access 24/7.

Good point. Hrm. "Needs"? No, I don't think so. It could be more or less. You can queue up 4 or 5 days worth.

I... don't think we need to include the cost of an internet connection.

First, it's more or less trivial. It's trivial value, for sure. In cost, nearly so. Well, maybe not. I live in Canada, and for most of North America, we pay flat rates. Flat rates for broadband, flat rates for dialup.

But in Europe, I know people often pay per minute for telephone use, and it can be horrendously expensive. So if you're on dialup, yep, big cost. I've never heard of bandwidth bills for ISPs at the home level though.

Second, we're most interested in added cost, not total cost. We want to know the difference between not running SETI, and running SETI. In the methodology I suggested below, we either cheat and assume 100% of the costs are added, or, if we can narrow it down, estimate the difference and compensate.

In the case of an internet connection though, I can't really imagine a case where it's ever an added cost (except on dialup in Europe with per-minute charges). While an internet connection costs money, I think it's safe to assume there's zero added cost, as opposed to assuming all of it.

How an internet connection might be added cost:

1) Per minute dialup charges, (as above).
2) If one was required to upgrade to a new connection type because of SETI, (from dialup to broadband) or from no net connection to some net connection. (And if dropping SETI, would return back to the lower level).
3) If one was required to upgrade to a more expensive package because of SETI, (from a 2Mb to a 4Mb line, because the annoyance of SETI was too much).

All of the above I consider to be trivial.

And then there is the cost of air conditioning, almost year round for me, to remove the heat generated by Setiathome computers.

Quite right. Almost impossible to estimate, but, quite right. My team captain is limited on how many machines he can hook up, without upgrading his air conditioner.

Two components:

1) Power consumption of the air conditioner.
2) Cost of purchasing/upgrading an air conditioner.

I think how electricity works, is you can directly conver the amount of watts something consumes, into raw heat. I.E. A computer using 400 watts is exactly the same as a 400 watt heater. A heater is just an easy way of throwing away electricity.

An air conditioner then, needs to do something to combat 400 watts of heat, in the above case. I think you might just be doubling your power cost right there, plus some extra because compressors aren't very efficient.

On the flip side.. I'm in Canada. In a basement. It's cold. There is no time it could not use a little more heat (and if it did, I could just open a window). While electrical heating is vastly more expensive than the alternative (natural gas central heating).. for whatever it's worth, it does heat some.

It's interesting to think about, but on the whole... this would be very hard to accurate estimate and include in our calculations. I think we could probably just assume the climate effect up or down about equals out, or nearly enough so. It really only rears its head when one specific users runs a *whole bunch* of computers... and that number is low. I.E. The cooling cost of 1 person with 100 computers is huge. The cooling cost of 100 people with 1 computer each is not noticible to any of them.

Good to get some feedback though. Good thinking.

---

Are we forgetting the cost of an Internet connection, or at least a share of the cost as only Setiathome needs access 24/7. And then there is the cost of air conditioning, almost year round for me, to remove the heat generated by Setiathome computers. I figured out that my star would only cost me about half a month's expense of running setiathome for them free of charge, so why not give some directly.

As with the power to run my machine, I'm paying for the internet connection anyway, and the connection is not __that__ busy.

This is the original idea behind SETI Classic -- we're using resources that are otherwise going unused.

---

As with the power to run my machine, I'm paying for the internet connection anyway, and the connection is not __that__ busy.

This is the original idea behind SETI Classic -- we're using resources that are otherwise going unused.

The idea of using otherwise "unused" resources is correct if we talk about unused CPU cycles. However, that makes it sound like it doesn't cost anything to run, which is untrue. Full use of the processor cost electricity. About 50% higher for my desktops and 250% more for my laptop. So, I'd only be paying 2/3rds as much if I didn't run boinc, more if I shut them off, and a whole lot more if I hadn't bought the AMD64's just to run boinc. LOL