I looked recently at a Dell intel core 2 extreme QX9650 3GHZ 1333FSB 12 MB L2 cache 4GB dual channel DDR2 SDRAM 800MHz, yada, yada...(or maybe yum, yum) but the price (though I can afford it) was ludicrous $2400 US. If I could build one like that for half that price then we'd be getting some where? Any how-to books on this stuff?
Ages and ages ago, when the IBM System /360 initial models came out, they were constrained in price and speed to fit Grosch's Law, which held that the computing power should increase as the square of the price.
That has been not true for a very, very long time. In desktop PC's these days you are almost certain to find the speed vs. purchase price optimum well below $1000. The speed power optimum is often at a somewhat higher point, as you amortize the power "wasted" by peripherals that don't actually compute, but still often below $1000.
However, the higher speed spec variants of what is fundamentally the same chip are often run at appreciably higher CPU voltages (insanely high in the hands of some overclockers). This can greatly increase the CPU power consumption.
If you actually consider acquiring a modern PC with the intent of retiring one (or more) old power-hogs just to run Einstein more efficiently, the trick is to avoid anything that hints of "gamer" orientation. High-end graphics cards aimed at them use fabulous amounts of power--you don't need it. The purchase price for higher-speed chips in the same line generally rises much faster than the performance, and the power consumption faster still.
Einstein (or any BOINC ap), takes far, far better use of multiple cores than does almost any consumer application. So the one "premium" thing you should value in such an economy-oriented choice is multiple cores. Last year's hot ride for build-it-yourself people aiming at good BOINC performance per unit purchase cost was the Intel Conroe-generation Q6600 (Kentsfield, though not many call it that). The Penryn-generation counterpart, Yorkfield, again a quad-core model, gets somewhat more performance per Hz from logic design improvements, and considerably lower idle power and somewhat lower operating power at a given frequency from design and process improvements. For something running 24x7 BOINC, systems based on it are likely cheaper than Q6600 systems in the long run because of power savings if you found a vendor who only charged something close to the price difference of the chips (well under $100 at retail).
However, looking at currently offered Dell desktop PC's aimed at home use, one sees some really, really inexpensive Q6600-using models, while the cheapest Penryn-generation offering they show is $600 more expensive (no, the CPU markup is not that high, they bundle it with more expensive graphics and such).
Ah, not actually true: if you start configuring the $499 Inspiron 530 they offer a CPU upgrade from Q6600 to Q9300 (the bottom Yorkfield) for a much more reasonable $100.
So right now Dell offers a Q6600 box, with plenty of hard drive and memory, an optical drive, keyboard, mouse, and Windows Vista Home Premium (but no monitor) for $499, plus $35 shipping and sales tax. No, it is not so fast as the monster you looked at, but vastly superior to it in BOINC output per purchase dollar, and quite likely superior to it in BOINC output per watt. Paying the extra $100 to go to the Q9300 would change that to almost certainly superior in BOINC output per watt.
Quite likely you can find something a bit cheaper that is just as fast from someone else, but you'd have trouble saving serious money at the performance level building it yourself. I see you currently run Windows, so I'll not push you to save yet more money (and currently on Einstein get more speed) by going for Linux.
wow! 126,905.00 secconds. or 35 hours, if i did my math right, so to do the other 26 wu's i need 37.91 days? i had to have done that math wrong :|
Looks about right to me, unless you have system down time or competing applications reducing you appreciably below 1 CPU second for Einstein per wall clock second, in which case it will be longer.
But they won't keep you busy for 38 days, deadline is long before then.
Ages and ages ago, when the IBM System /360 initial models came out, they were constrained in price and speed to fit Grosch's Law, which held that the computing power should increase as the square of the price.
LOL...
Grosch's Law!!?? I hate when I get forced into dating myself! ;-)
However, that always struck me as marketing crapola as it was typically applied. :-)
Anyway, all good advice in there, but if we keep up at this rate we'll take all the sport out of it for a first time builder. :-D
wow! 126,905.00 secconds. or 35 hours, if i did my math right, so to do the other 26 wu's i need 37.91 days? i had to have done that math wrong :|
Looks about right to me, unless you have system down time or competing applications reducing you appreciably below 1 CPU second for Einstein per wall clock second, in which case it will be longer.
But they won't keep you busy for 38 days, deadline is long before then.
Agreed, but TDCF should have made the big jump up. So I'd just start trimming about half of the R4's now, starting with the 'newest' ones to avoid having the template go into limbo as datapack group members move on in a week or so.
Actually, it's not that bad. You have control of all of your money and you don't feel guilty for taking yourself out for dinner and drinks, and whatever the rest of the night entails... ;-)
Agreed, but TDCF should have made the big jump up. So I'd just start trimming about half of the R4's now, starting with the 'newest' ones to avoid having the template go into limbo as datapack group members move on in a week or so.
Alinator
err. meaning the ones that have a deadline of 9/13 as apposed to 9/11 ? i.e. bottom up?
seeing without seeing is something the blind learn to do, and seeing beyond vision can be a gift.
Actually, it's not that bad. You have control of all of your money and you don't feel guilty for taking yourself out for dinner and drinks, and whatever the rest of the night entails... ;-)
LOL...
Badah, bump... That was bad! :-)
Reminds me of a joke about girlfriends and ten pin bowling.... But I don't think I want to go there! :-D
The only wildcard is if you don't have a copy of Windows to install. Then you will take somewhat of a hosing on that, since you won't get the sweetheart discount deal you'd get from an OEM. However, there's always Linux or eBay to ease that pain.
Hmm, I tried installing eBay 1.1 but now I'm waiting for eBay Service Pack 1...
:D
RE: I looked recently at a
)
Ages and ages ago, when the IBM System /360 initial models came out, they were constrained in price and speed to fit Grosch's Law, which held that the computing power should increase as the square of the price.
That has been not true for a very, very long time. In desktop PC's these days you are almost certain to find the speed vs. purchase price optimum well below $1000. The speed power optimum is often at a somewhat higher point, as you amortize the power "wasted" by peripherals that don't actually compute, but still often below $1000.
However, the higher speed spec variants of what is fundamentally the same chip are often run at appreciably higher CPU voltages (insanely high in the hands of some overclockers). This can greatly increase the CPU power consumption.
If you actually consider acquiring a modern PC with the intent of retiring one (or more) old power-hogs just to run Einstein more efficiently, the trick is to avoid anything that hints of "gamer" orientation. High-end graphics cards aimed at them use fabulous amounts of power--you don't need it. The purchase price for higher-speed chips in the same line generally rises much faster than the performance, and the power consumption faster still.
Einstein (or any BOINC ap), takes far, far better use of multiple cores than does almost any consumer application. So the one "premium" thing you should value in such an economy-oriented choice is multiple cores. Last year's hot ride for build-it-yourself people aiming at good BOINC performance per unit purchase cost was the Intel Conroe-generation Q6600 (Kentsfield, though not many call it that). The Penryn-generation counterpart, Yorkfield, again a quad-core model, gets somewhat more performance per Hz from logic design improvements, and considerably lower idle power and somewhat lower operating power at a given frequency from design and process improvements. For something running 24x7 BOINC, systems based on it are likely cheaper than Q6600 systems in the long run because of power savings if you found a vendor who only charged something close to the price difference of the chips (well under $100 at retail).
However, looking at currently offered Dell desktop PC's aimed at home use, one sees some really, really inexpensive Q6600-using models, while the cheapest Penryn-generation offering they show is $600 more expensive (no, the CPU markup is not that high, they bundle it with more expensive graphics and such).
Ah, not actually true: if you start configuring the $499 Inspiron 530 they offer a CPU upgrade from Q6600 to Q9300 (the bottom Yorkfield) for a much more reasonable $100.
So right now Dell offers a Q6600 box, with plenty of hard drive and memory, an optical drive, keyboard, mouse, and Windows Vista Home Premium (but no monitor) for $499, plus $35 shipping and sales tax. No, it is not so fast as the monster you looked at, but vastly superior to it in BOINC output per purchase dollar, and quite likely superior to it in BOINC output per watt. Paying the extra $100 to go to the Q9300 would change that to almost certainly superior in BOINC output per watt.
Quite likely you can find something a bit cheaper that is just as fast from someone else, but you'd have trouble saving serious money at the performance level building it yourself. I see you currently run Windows, so I'll not push you to save yet more money (and currently on Einstein get more speed) by going for Linux.
wow! 126,905.00 secconds. or
)
wow! 126,905.00 secconds. or 35 hours, if i did my math right, so to do the other 26 wu's i need 37.91 days? i had to have done that math wrong :|
seeing without seeing is something the blind learn to do, and seeing beyond vision can be a gift.
RE: wow! 126,905.00
)
Looks about right to me, unless you have system down time or competing applications reducing you appreciably below 1 CPU second for Einstein per wall clock second, in which case it will be longer.
But they won't keep you busy for 38 days, deadline is long before then.
RE: Ages and ages ago, when
)
LOL...
Grosch's Law!!?? I hate when I get forced into dating myself! ;-)
However, that always struck me as marketing crapola as it was typically applied. :-)
Anyway, all good advice in there, but if we keep up at this rate we'll take all the sport out of it for a first time builder. :-D
Alinator
RE: RE: wow! 126,905.00
)
Agreed, but TDCF should have made the big jump up. So I'd just start trimming about half of the R4's now, starting with the 'newest' ones to avoid having the template go into limbo as datapack group members move on in a week or so.
Alinator
RE: I hate when I get
)
Actually, it's not that bad. You have control of all of your money and you don't feel guilty for taking yourself out for dinner and drinks, and whatever the rest of the night entails... ;-)
RE: Agreed, but TDCF
)
err. meaning the ones that have a deadline of 9/13 as apposed to 9/11 ? i.e. bottom up?
seeing without seeing is something the blind learn to do, and seeing beyond vision can be a gift.
Correct. Although it's
)
Correct. Although it's sometimes hard to determine which is which on multiple DL events without looking at the UNIX timestamp.
In any event, this is one of those cases where close is good enough. ;-)
Alinator
RE: RE: I hate when I
)
LOL...
Badah, bump... That was bad! :-)
Reminds me of a joke about girlfriends and ten pin bowling.... But I don't think I want to go there! :-D
Alinator
RE: The only wildcard is if
)
Hmm, I tried installing eBay 1.1 but now I'm waiting for eBay Service Pack 1...
:D