If you haven't seen this message then I reckon it's well worth a read ;). Briefly, Akos proposed a formula for working out how many 'standard' results could be crunched for a buck, taking both capital and running costs into account. As many people eventually get well and truly 'bitten', there's always a need to remind yourself about the cost of crunching. My purpose in starting this thread is to suggest a little light-hearted competition to see who has the highest 'Akos' computer based on the standard application only.
Have a look at Akos' formula. I'm going to propose two very minor modifications, one of which will boost your 'Akos' value - so no cheating now!!! :).
Firstly let's define a reference result as one that returns 100 credits. So if your machine is currently doing long results and returning 178 credits say, you are currently processing a 1.78X ref result. If you are doing shorts returning 19.7 credits, you are doing a .197x ref result.
Secondly we need to have a standard way of putting a value on the machine you are using. This project is going to run for about a year so the sensible thing is to take the value of your computer now and subtract what you reckon its residual value will be in a year's time. Obviously you could cheat by claiming your machine doesn't lose value over a year :). Below is a copy of Akos' formula with the suggested changes.
I can suggest this form: e = r / t / ( p * r * c + s )
e = 'Akos' value of your machine in units of WUs / USD
p = power consumption ( kW )
s = system value change over 1 year ( USD )
c = cost of electricity ( USD / kWh )
t = time of reference WU ( hour / WU )
r = run-time of the project ( hour )
- p ~ Measure it if possible or estimate based on components being used or a sensible fraction of your PSU rating. Show your reasoning for peer review.
- c ~ Use what is on your electricity account
- t ~ Calculate = Crunch time (secs) / 3600 / credits awarded * 100
- r ~ 1 year = 365.25 days = 8766 hours
So here are a couple of examples taken from some of my machines:-
HP e-PC PIII 933MHz @ 1015MHz/256MB/6.4GB. It uses a laptop style power supply whose max rating is 19V x 3.16A = 60 watts. I'm guessing it is probably not fully maxed out so I'll use 50 watts = .05KW
On my latest elec bill, my cost is 0.111 USD/KWH
The machine cost USD50 two months ago and I'm sure I could get it all back in a years time but I'll depreciate it by USD20.
The machine is currently crunching long units in 81,042 secs for 174.63 credits. So t = 81042 / 3600 / 1.7463 = 12.89 hours / ref WU.
Akos = 8766 / 12.89 / (0.05 * 8766 * 0.111 + 20) = 9.9 Akoses
Athlon 64 3000+ @ 2320MHz/512MB/20GB. It's in a mini ATX case with a 200W PSU. I reckon it's probably drawing about 140 watts - it feels about three times as "hot" as the e-PC :). I built it recently from a mix of new and 2nd hand components for about 220 USD. After a year I reckon it will still be worth 150 USD so its value loss is 70 USD.
It does long results in 29,800 seconds returning 176.5 credits so t = 29,800 / 3600 / 1.765 = 4.69 hours / ref WU.
Akos = 8766 / 4.69 / (0.14 * 8766 * 0.111 + 70) = 9.06 Akoses
Quite surprising actually. My gut feeling was that the AMD64 would win because it was quite fast and built to a budget with not really that much depreciation being allowed for.
It so happens that this machine did some results using the 0712 patch a week ago. Here is the calc for t = 22,580 / 3600 / 1.775 = 3.53 hours / ref WU.
Akos = 8766 / 3.53 / (0.14 * 8766 * 0.111 + 70) = 12.04 Akoses
It's a pretty good non-emotional argument to feed back to Bruce as he ponders the fate of the Akos optimisations.
EDIT: Added highlighting
EDIT2: So come on you lot!! Show us all your Akos ratings. Who has got the biggest one!! :).
Remember - standard app only unless clearly identified otherwise and document your assumptions :).
____________
Cheers,
Gary.
Here is another machine - a low end P4
HP Vectra VL420 P4 1.6GHz @ 2.0GHz/256MB/20GB. Its power supply is rated at 185W and it is overclocked so I'm going to assume that it's using about 120 watts even athhough there is nothing but the mobo, cpu, RAM and HDD being powered. I bought a bunch of these recently at a Govt auction fully working for USD70 each. Similar machines sell on eBay for about USD100 to USD120 and I expect to be able to sell them in a year's time for at least what I paid for them, perhaps more. However I'll be conservative and assume I only get USD50 each so the value loss I'll use is USD20.
It's doing long results in 64,700 seconds returning 178.12 credits so t = 64,700 / 3600 / 1.7812 = 10.09 hours / ref WU.
Akos = 8766 / 10.09 / (0.12 * 8766 * 0.111 + 20) = 6.35 Akoses
Very interesting.... I hadn't worked this out before starting this message and I thought I'd be posting a higher value than either of the two previous ones and not a lower one. I knew the high end PIIIs were more efficient than the low end P4s but I didn't think it would show up so clearly. I really did expect the P4s to win here. It just goes to show how sensitive the calculation is to actual power consumption so I guess I'd better invest in a power meter and actually measure it rather than just guessing. However even if I assume the PIII is using the full rated 60W of the power pack and the P4 is only using 100W, the PIII still wins by 8.68 to 7.41. Quite surprising!!
There is another auction tomorrow with a bunch of HP e-PCs listed that should go pretty cheaply. I just might go and get 'em all and sell these P4s while they are still bringing a bit of a premium :).
EDIT: It's a bit of a sad indictment of the P4 Williamette cores that one running at 2.0GHz takes 10.09 hours for the ref WU whilst the PIII at just over half the speed only takes a small amount more - 12.89 hours. Maybe the Govt really knew what it was doing by flogging these off for only 70 bucks apiece :).
____________
Cheers,
Gary.
Hi Gary!
I think you can get a bit more comparable results if you drop out the 'price of the configuration' value from the evaluation ( but of course your result will not consits an important iformation ). And you should define the WU (eg.: try to upload for us a zipped boinc directory with one WU).
edit: You should also use constant values for the 'price of electricity' and the 'project time'... There would be only two variables ( power consumption and WU time ). These values are easily measurable.
...And you should define the WU (eg.: try to upload for us a zipped boinc directory with one WU).
Hi Akos,
This is meant to be a bit of fun and not to be taken too seriously :). I'm not pretending that the values are rigorously determined :).
Because I have 25 HP e-PCs for example, all running between 800 to 1100 MHz with the bulk of them around 950MHz, I've got a good opportunity to see what sort of variation there is between different long results on comparable hardware. Yes, I do see examples where the time varies a bit on the one frequency. Yes, I do see examples where there are sudden changes in the credit awarded when the crunch time has in fact gone in the opposite direction. However, by and large, the credit awarded does seem to be a good enough approximation to the work content and therefore to the time that should be taken on equal hardware. Certainly close enough for the purposes of a fun exercise like this. All of those 25 e-PCs are producing times roughly consistent with their actual speed and with the work content as approximated by the granted credit.
I'm sure there are bigger errors introduced by assuming rather than measuring the power consumed. So to keep this as simple as possible so that people will be encouraged to enter into the spirit of the 'contest', I'm quite happy to just use the credit granted as the yardstick, and accept that the answers will be a bit 'fuzzy'. Also if people want to leave out the depreciation term and just produce a figure based on electricity alone, that's fine with me. Part of the reason for me starting this thread was to get a better handle myself on a "running cost figure of merit" for different types of systems irrespective of the capital costs involved. I need to thank you for getting me started on that as I've already made some interesting discoveries :).
Of course, if someone volunteers to capture a few reference results and to organise a distribution and recording system to keep track of it all, I would certainly join in. However I know my limitations and it ain't gonna be me :).
____________
Cheers,
Gary.
____________
Hi Gary!
I think you can get a bit more comparable results if you drop out the 'price of the configuration' value from the evaluation
I agree, the less subjective the results the better.
Give me a bit and I'll have values for oveverclocked and stock results.
Dave
____________
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
A reasonable depreciation estimate is 1/3rd per year. This is what the Aussie Tax Office uses anyway. For business purposes and tax claims a computer becomes worthless, 'on the books', in 3 years. I think this is the steepest possible of all items on a depreciation schedule. To compare: generally a car will 'write down' over 5 years and a filing cabinet over 12. Any accountants out there care to comment?
Of course we use them for longer. But I must admit to replacing at least one component or other within that time period for any box I've owned. This is generally an upgrade, but also for outright failure. I don't include items outside the box here, and I won't even start on printers ... :-)
Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
A reasonable depreciation estimate is 1/3rd per year. This is what the Aussie Tax Office uses anyway. For business purposes and tax claims a computer becomes worthless, 'on the books', in 3 years. I think this is the steepest possible of all items on a depreciation schedule. To compare: generally a car will 'write down' over 5 years and a filing cabinet over 12. Any accountants out there care to comment?
Of course we use them for longer. But I must admit to replacing at least one component or other within that time period for any box I've owned. This is generally an upgrade, but also for outright failure. I don't include items outside the box here, and I won't even start on printers ... :-)
Cheers, Mike.
So if a box is more than 3 years old is its depreciation value $0.00? I've got a PII 450 crunching away, and there is no way it was worth anything at the beginning of the project, and it certainly wont be worth jack by the end of S5.
____________
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
So if a box is more than 3 years old is its depreciation value $0.00? I've got a PII 450 crunching away, and there is no way it was worth anything at the beginning of the project, and it certainly wont be worth jack by the end of S5.
Yeah, zero after 3 years ( from new ) is the value. So you get to deduct 1/3rd the original price from your taxable income, for each of the following three years from purchase. I am unsure as to how much this applies if the asset is sold on ie. from second hand. This requires that the item is used to generate business income ie. is a legitimate expense.
I think the 1/3 rule reflects the speed of change in the IT industry - obsolescence is quite rapid - and the fact that the politicians like the deduction. As my accountant told me long ago: arrange your tax matters alongside the pollies, as they'll never crap in their own nest! How true ..... :-)
Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
I think the 1/3 rule reflects the speed of change in the IT industry - obsolescence is quite rapid - and the fact that the politicians like the deduction. As my accountant told me long ago: arrange your tax matters alongside the pollies, as they'll never crap in their own nest! How true ..... :-)
I believe the figure is 30% p.a. here—but I’m not sure how it’s applied. It would seem more logical to reduce the value proportionally each year (like compounding interest) so that it never disappears: 34% of the original amount after three years, 17% after five, 3% after ten.
Software (other than an OS) depreciates at 100% p.a., so has zero book value after a year.
____________
I believe the figure is 30% p.a. here—but I’m not sure how it’s applied. It would seem more logical to reduce the value proportionally each year (like compounding interest) so that it never disappears: 34% of the original amount after three years, 17% after five, 3% after ten.
I think the phrase 'Australian Tax Office' and the word 'logical' have yet to appear in a semantically correct sentence! :-)
Software (other than an OS) depreciates at 100% p.a., so has zero book value after a year.
Well we do love the immediate write downs of course .... :-)
Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal