How to efficiently crunch SETI

[Roisen]

When running a computer that is dedicated to SETI@home, folding@home, or any other DC project, the goal is to reduce the power consumption to as little as possible. This will not only directly reduce your power costs by drawing less energy, but will also produce less heat, which may save you money as well. Note that computers are not designed to be heaters, and it is more energy efficient to reduce your computer's heat output even if you live in a cold part of the world. Heaters have a much higher heat/watt output than computers do.

The following are steps that you can take to reduce the amount of heat that your computers produce, and energy that they use:



Run your computers headless
Running your computers headless means without a monitor, keyboard, or mouse. If changes need to be made to the computers configuration, they can be done by Remote Desktop over the network.

Remove any unused peripherals
Soundcards, 56k modems, and video adapters all use energy when they're plugged in. Also, you can disable unused onboard peripherals through the device manager in windows, such as firewire, usb, unused network adapters, COM port, and unused IDE/ATA channels. Netword adapters that are in use may have a "power saving mode". Use it if it is available. Turn off your monitor when leaving the room for an extended period of time (thanks Dk Jedi Allianc).

Overvolt intelligently(thanks alexhk)
""Power consumption scales linearly with respect to clock speed. But when you increase the voltage, power consumption increases with the square of the voltage.

i.e. - A 10% increase in core voltage will result in a 21% increase in power consumption.
...
While performance per watt out of one's CPU is important there is quite a bit of overhead involved in running a rig that can't be avoided. I'm speaking of the power needed for motherboard, RAM, hard drive, ect. I can envision a situation where say 4 more heavily overclocked machines (were overvolted) resulted in more RAC and less power consumption than 5 boxes that were for the most part left alone. I guess the key would be to work out power consumption of the CPU relative to that of the rest of the system and find the sweet spot.""

Buy more efficient parts (thanks Dk Jedi Allianc)
""If you intend to crunch SETI only, better to go with 2 C2D than 3-4 AMD X2. Intel rule for SETI.
If you purchase an AMD X2, go after as much L2 cache as possible.""
Look for power supplies with a higher efficiency rating.

Change your DC client's preferences
For BOINC, set the preferences to write to disk every xx seconds. A larger time will utilize the hard disc less, and thus consume less power, however a computer or power failure will mean more work lost.


Use laptops(thanks freakdiablo)
""Use laptops when possible. 3 of the 4 computers I have running are IBM laptops, and each one only uses 65w, and thats including a core duo T2300 that puts out 950rac.""
(laptops are designed with energy conservation in mind. The fewer watts they draw, the longer the battery will power it, and battery life is something that many, if not all customers look for)



I hope this is helpful! Feel free to post new ideas below and I will add them in up here (with credit to you of course!)





_

 

[Dk Jedi Allianc]

Turn off your monitor when leaving the room.

If you intend to crunch SETI only, better to go with 2 C2D than 3-4 AMD X2. Intel rule for SETI.
If you purchase an AMD X2, go after as much L2 cache as possible.

 

[QuietIce]

I might point out, since we're running through everything to save power, that water cooling systems tend to use much more power than air cooling. Even a small pump like the MCP350 is rated at 9W - that'll run a lot of fans. If you can cool 3 or more computers with one water-cooling loop then you can make water cooling efficient enough to compete with fans.


I have my doubts about turning off the monitor. While I'm sure sleep mode uses a tad more power than "off" I'm also sure you would shorten the life of your CRT by going whole hog with this. Turn it off over-night? Sure - once a day shouldn't be a problem for the average CRT. Turn it off 3-4 times a day? You'll send it to an early death. I should point out this only applies to CRTs - I have no clue how well LCDs are built to handle this.


With 2 exceptions I NEED all my computers - so it's AMD for me. I have enough trouble spending 25% of a computer's budget just to buy XP Pro ...

 

[JamesXP]

Tut, water cooling = higher overclocks = higher income.

 

[QuietIce]

Tut, water cooling = higher overclocks = higher income.

You'll notice my sig? I'm not trading in my pumps for an Ultra-120! But this thread was about efficiency - water cooling isn't as efficient as air cooling. And the OP specifically mentions OC'ing as something to avoid.

I'm not sure I agree with this philosophy or not, I'd need to run some tests and do some research to refute it, though, as I've no data to say otherwise. It's obvious on the face of it that OC'ing uses more energy but whether it's less efficient or not I can't say. Do you have any data one way or the other? I'm up for a long-term discussion of OC'ing v efficiency - and I'd be on your side, too ...!

 

[freakdiablo]

Use laptops when possible. 3 of the 4 computers I have running are IBM laptops, and each one only uses 65w, and thats including a core duo T2300 that puts out 950rac.

 

[QuietIce]

Not a bad idea - but laptops are a little expensive for my taste.

However, a mobile processor in a uATX board w/built-in video (or old, used PCI video) is a good idea! Still great processing power w/the cost of a laptop or the overhead of a desktop ...

 

[freakdiablo]

Not a bad idea - but laptops are a little expensive for my taste.

I know, guess Im a bit lucky my dad works IBM. He was able to get my moms (R60) for I think he said ~$500, about the price of a low end desktop.

 

[Dk Jedi Allianc]

I have my doubts about turning off the monitor. While I'm sure sleep mode uses a tad more power than "off" I'm also sure you would shorten the life of your CRT by going whole hog with this. Turn it off over-night? Sure - once a day shouldn't be a problem for the average CRT. Turn it off 3-4 times a day? You'll send it to an early death. I should point out this only applies to CRTs - I have no clue how well LCDs are built to handle this.

I got one Phillips 17" CRT that are 7 years old. I bougth during the classic SETI. And I used to turn it on and off like 10 times per day. Today my 11-year old boy uses this to play games on...

Also when I wrote leave the room I meant if leaving for hours, not just to go and get a cold beer...

 

[Roisen]

I'm not sure I agree with this philosophy or not, I'd need to run some tests and do some research to refute it, though, as I've no data to say otherwise. It's obvious on the face of it that OC'ing uses more energy but whether it's less efficient or not I can't say. Do you have any data one way or the other? I'm up for a long-term discussion of OC'ing v efficiency - and I'd be on your side, too ...!

There is a sticky about overvolting somewhere in these forums that explains it, also experiance should tell you that overclocking is much less effecient (after a point) for example, to get from 2.13Ghz to 3.00Ghz, I didn't need any voltage bumps from the stock 1.21, but to get from 3.00Ghz to 3.50Ghz, I needed to volt over 1.4, and higher voltages at the same amperages = higher wattage throughput. Of course I have all of my non-OEM computer overclocked, but like you said, this is aimed more at the guys who constantly have their wives on them to turn off some of their computers because it's too hot.

I have my doubts about turning off the monitor. While I'm sure sleep mode uses a tad more power than "off" I'm also sure you would shorten the life of your CRT by going whole hog with this. Turn it off over-night? Sure - once a day shouldn't be a problem for the average CRT. Turn it off 3-4 times a day? You'll send it to an early death. I should point out this only applies to CRTs - I have no clue how well LCDs are built to handle this.

Taken from a University of Waterloo webpage (link):

A Sony 17" monitor, by far our most common, consumed 75w when in use. When power-saver mode kicks in (and the monitor goes black with a yellow indicator light) the power consumption is negligible to the point that our test equipment did not even register any power use.

Of course they don't mention what test equipment they use, so "negligible" could be as much as 5 watts, which in the scope of this thread, I feel is a significant amount.
I would also like to see some proof that CRTs endure that kind of wear during startup. I guess it's my turn to straddle the fence.

 

[eaglescouter]

My boxes are mostly headless.

My boxes that do have monitors; the monitor is powered off unless I am sitting there using it.

 

[Alexhk]

On the overclocking and overvolting topic:

Power consumption scales linearly with respect to clock speed. But when you increase the voltage, power consumption increases with the square of the voltage.

i.e. - A 10% increase in core voltage will result in a 21% increase in power consumption.

Hence the diminishing returns argument that Roisen is putting forth. I like the efficiency minded thinking but I'm not sure its so cut and dry that one could make blanket statement like, never overvolt.

While performance per watt out of one's CPU is important there is quite a bit of overhead involved in running a rig that can't be avoided. I'm speaking of the power needed for motherboard, RAM, hard drive, ect. I can envision a situation where say 4 more heavily overclocked machines (were overvolted) resulted in more RAC and less power consumption than 5 boxes that were for the most part left alone. I guess the key would be to work out power consumption of the CPU relative to that of the rest of the system and find the sweet spot. Little late to do out any math so hopefully someone gets my point, hehe.

 

[Roisen]

On the overclocking and overvolting topic:

Power consumption scales linearly with respect to clock speed. But when you increase the voltage, power consumption increases with the square of the voltage.

i.e. - A 10% increase in core voltage will result in a 21% increase in power consumption.

Hence the diminishing returns argument that Roisen is putting forth. I like the efficiency minded thinking but I'm not sure its so cut and dry that one could make blanket statement like, never overvolt.

While performance per watt out of one's CPU is important there is quite a bit of overhead involved in running a rig that can't be avoided. I'm speaking of the power needed for motherboard, RAM, hard drive, ect. I can envision a situation where say 4 more heavily overclocked machines (were overvolted) resulted in more RAC and less power consumption than 5 boxes that were for the most part left alone. I guess the key would be to work out power consumption of the CPU relative to that of the rest of the system and find the sweet spot. Little late to do out any math so hopefully someone gets my point, hehe.

You are right. Will change my main post.

 

[QuietIce]

On the overclocking and overvolting topic:

Power consumption scales linearly with respect to clock speed. But when you increase the voltage, power consumption increases with the square of the voltage.

i.e. - A 10% increase in core voltage will result in a 21% increase in power consumption.

Hence the diminishing returns argument that Roisen is putting forth. I like the efficiency minded thinking but I'm not sure its so cut and dry that one could make blanket statement like, never overvolt.

While performance per watt out of one's CPU is important there is quite a bit of overhead involved in running a rig that can't be avoided. I'm speaking of the power needed for motherboard, RAM, hard drive, ect. I can envision a situation where say 4 more heavily overclocked machines (were overvolted) resulted in more RAC and less power consumption than 5 boxes that were for the most part left alone. I guess the key would be to work out power consumption of the CPU relative to that of the rest of the system and find the sweet spot. Little late to do out any math so hopefully someone gets my point, hehe.

Unfortunately, that's only half of the equation, the other half is the heat load to the environment. While this is nothing in the winter (maybe even a plus!) or in areas where the temperature is not controlled, it's a big part of it if your rigs are running inside a central air home. One would have to look at the added thermal load to the room for a whole rig and then figure the extra load due to OC and over-volting. That's the part where a lot of testing and research would have to be done b/c now you're working with A/C efficiencies, power costs, etc. - in other words a royal PITA unless you have some experience in those areas. And power costs vary a lot from area to area. Here in the mid-west many states are over-powered and electricity is cheap. As I understand it, on the coasts it's a much different story.

Putting the crunchers in the basement might help some but the heat still has to go somewhere. Maybe the most efficient thing you can do IS water-cooling - using a geo-thermal radiator. Still, that takes a rather large pump to implement and we are only talking about maybe 4 months of the year, but you could cool multiple computers with it. That puts us back to testing and research to answer our question ...