Math for Farmers

Message boards : Number crunching : Math for Farmers

To post messages, you must log in.

AuthorMessage
The_Bad_Penguin
Avatar

Send message
Joined: 5 Jun 06
Posts: 2751
Credit: 4,271,025
RAC: 0
Message 28823 - Posted: 2 Oct 2006, 19:30:41 UTC
Last modified: 2 Oct 2006, 19:40:07 UTC

Been watching a few of the recent posts, and some of their links over to other projects, re getting a small crunching farm up and running.

It seems the most valuable piece of information that I have gleamed from all of this is to get to the mailbox first, to get the electric bill before your significant other does!

Given that it is not difficult to obtain cheap/free "older" pc's in the 1 GHz range (I expect this will certainly be true after Jan 2007 with Vista and AMD/Intel's latest and greatest), it seems the "real" cost is feeding them.

Unfortunately, I live in an area with one of the highest electric rates in the country.

It wouldn't seem to make sense to accept 10 free P3 1GHz boxes, or purchase a cheap eBay special on a Quad Xeon 700MHz server when, in the long run it would be "cheaper" to run X2's or Core Duo's.

How do YOU decide how many crunchers is enough, and what are the minimum specs you would farm with?

Any tips or tricks for cutting back on power consumption?

Thoughts and comments appreciated.
ID: 28823 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Feet1st
Avatar

Send message
Joined: 30 Dec 05
Posts: 1755
Credit: 4,690,520
RAC: 0
Message 28824 - Posted: 2 Oct 2006, 19:35:29 UTC

Can't say I've tried it, but my main thought for reducing operating costs is just to plan on some productive use for the heat output of the machines. And NOT to run them during Summer when you are running air conditioning of the same room.

If your electic bill goes up, but your heating bill goes down, then your net costs are reduced.
Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/
ID: 28824 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
BennyRop

Send message
Joined: 17 Dec 05
Posts: 555
Credit: 140,800
RAC: 0
Message 28826 - Posted: 2 Oct 2006, 20:44:56 UTC

If you can produce the same amount of work out of 1 (non top of the line) machine as 5-10 older machines - I'd opt for saving the space and the power.

One of the issues that came up in the past on another project was whether it was worth it to buy the newer energy efficient 462 cpus, rather than the older version that had a 5,10, or 20 watt difference. Someone provided a calculator for cost of electricity for year (KW/h charges are different around the country and around the globe) - and the cost of the new cpu was 2 or 3 years of electrical savings for me. I opted 6 months later for a cpu with 50% better performance with the same power use as the cpu it replaced.

If you're going to replace something due to power prices, and the upgrade cost is going to take 4 years to break even; and you replace the systems in 2 years anyway - then you'll have a hard time justifying the upgrade on energy price savings alone.

Compare your options, and see which makes more sense for your situation.
ID: 28826 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
The_Bad_Penguin
Avatar

Send message
Joined: 5 Jun 06
Posts: 2751
Credit: 4,271,025
RAC: 0
Message 28827 - Posted: 2 Oct 2006, 21:03:13 UTC - in response to Message 28826.  

Other than After-the-Fact seeing the increase(s) in the electric bill, how could I best guess-timate the additional monthly cost of electricity for different possible configurations/acquisitions of crunchers?

I want to get some additional boxes, no monitors or printers attached. Do I track down specs on the cpu and add in some extra watts for the HDD and NIC?

So, do I take a "Free" P3, or spend up to $XX on a P4 with twice the MHz?

Just need to know the variables, so I can make a simple Excel graph showing my various break-even points.

To keep it easy, assuming KW/h @ $0.25.

How much $ am I spending per month on power consumption, and am I maximizing the results (measured by BOINC points).

Someone provided a calculator for cost of electricity for year (KW/h charges are different around the country and around the globe)...

ID: 28827 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1830
Credit: 119,208,549
RAC: 2,517
Message 28839 - Posted: 2 Oct 2006, 22:57:34 UTC - in response to Message 28823.  

Any tips or tricks for cutting back on power consumption?

Boot media
You can run XP or linux from compactflash which means you don't have as high an initial outlay as you don't have to buy a HD and also reduced running costs as CF uses negligable power. I have a mediacentre running on CF and I just run Rosetta from a shared folder on my server.

Small changes like that are amplified as you then have reduced cooling requirements (especially on aircon if you use that), and reduced PSU requirements.

It's also silent, which is a bonus.

You can also net-boot linux, but I wouldn't be able to help there!

PSU
A high efficiency PSU can make quite a big difference. You can get PSUs that are up to around 90% efficient. There's a really good PSU site here which recommends the Seasonic S12 series - I've just got one and am impressed.

Other things I can think to consider for efficient computing are:

Overclocking increases throughput, but not necessariliy linearly with energy consumption. I expect it would probably hold that if you can overclock without increasing the voltage then you will get a good return for the slightly increased power consumption. Alternatley, you can keep the speed the same and drop the voltage, which has increased the battery life on my laptop by around 20%.

A CPU with a large cache will get more throughput as it doesn't have to keep fetching data from RAM.

Passive cooling is better than active if you can, but keeping components cool reduces their resistance and so reduces energy consumption, so some active cooling is usually wise.

Memory power useage (least to most): DDR2, DDR, SDRAM

No need for anything in the way of graphics cards. You can actually get by without one at all (using a dummy graphics driver) but I expect you'll save yourself a lot of hassle by just running either an old graphics card or onboard.

Run headless - use a KVM rather than leaving a keyboard and mouse plugged into each computer (and share a monitor between them).

Run remotes - get other people to install Rosetta on their computers - it's free!

HTH
Danny
ID: 28839 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
BennyRop

Send message
Joined: 17 Dec 05
Posts: 555
Credit: 140,800
RAC: 0
Message 28840 - Posted: 2 Oct 2006, 23:00:40 UTC
Last modified: 2 Oct 2006, 23:03:07 UTC

Some of the review sites have used a power meter to show the total system power draw. There's inexpensive power meters like this available out there. It'll show how much power each system you plan on using is eating up. Or show the power usage of all the other items in the house. My Apple II+ was blamed for a huge spike in power usage which turned out to be a dying freezer.

According to this page, a 90nm Athlon 64 3500+ cpu based system used 151 watts; a 130nm Athlon 64 3500+ cpu based system used around 180 watts, and a 90nm Prescott P4 3.4Ghz cpu based system used around 236 watts. In full cpu usage mode.

90nm Athlon 64 = 0.151 KW*24 hrs/day * 365.25 days/year * $0.25/KWhr = $330.91 per year.
130nm Athlon 64 = 0.180 KW*24 hrs/day * 365.25 days/year * $0.25/KWhr = $394.47
90nm Prescott = 0.236 KW*24 hrs/day * 365.25 days/year * $0.25/KWhr = $517.19

To compare the cpu power usage, there's this chart: here.
Although that doesn't take into consideration the higher power usage of motherboard components of modern north bridges, the change in power usage for different ram configurations, and the higher power usage for modern video cards.

And keep in mind that for the Athlon systems above, the cpus were listed as being "less than 89 Watts" (130nm) and "less than 65 watts" (90nm); so over half the power draw of the 2 listed Athlon systems is for things other than the cpu.

Perhaps someone else can add in the power usage of a few systems in the P3 600-1Ghz range. Or some of the newer, lower power wasting Intel cpus.

Combine this with something like Tony's older credit/hour lists; only using a week's worth of the new credits - and we'd be able to see the sweet spot for KW/credit.
ID: 28840 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
The_Bad_Penguin
Avatar

Send message
Joined: 5 Jun 06
Posts: 2751
Credit: 4,271,025
RAC: 0
Message 28841 - Posted: 2 Oct 2006, 23:01:55 UTC - in response to Message 28839.  
Last modified: 2 Oct 2006, 23:02:10 UTC

Now that's the best suggestion I've heard yet for reducing power consumption!

Other suggestions are being duly noted.

Thanx.

get other people to install Rosetta on their computers - it's free!

ID: 28841 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1830
Credit: 119,208,549
RAC: 2,517
Message 28842 - Posted: 2 Oct 2006, 23:05:55 UTC - in response to Message 28827.  

So, do I take a "Free" P3, or spend up to $XX on a P4 with twice the MHz?

If you want efficient crunchers then I don't think P4s are the way to go. They're great at some stuff, but being efficient isn't one of them! If the P3s are Tualatin core (1GHz+) then they're pretty efficent CPUs. The older Coppermines (to 1GHz) aren't as efficient.

I expect that, in order of throughput per watt, CPUs would be in this order (most efficient to least):

Core2
Core
Pentium M (Dothan etc)
Athlon 64 (especially Venice core onwards)
Athlon XP
P3 (Tualatin)
P3 (coppermine & older)
Athlon
P4
ID: 28842 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
BennyRop

Send message
Joined: 17 Dec 05
Posts: 555
Credit: 140,800
RAC: 0
Message 28847 - Posted: 3 Oct 2006, 4:08:53 UTC

Some of the higher end hardware from Tom's Hardware
Intel Core 2 Extreme X6800
Gigabyte ATI Radeon X1900 XTX
Asus P5WDH Deluxe
2x Western Digital WD15000ADFD
DVD-ROM 16x
PC-Power&Cooling Turbo-Cool 510 SSI
Terratec 7.1 Sound

i.e. gaming setup.
Supposedly, the Athlon was setup with similar hardware. (Different motherboard, of course.)
Remove the soundcard, one of the Raided drives, and replace the 440ppd F@H Radeon x1900 xtx card to lower the power usage of the system..

Core2Extreme 6800: 217 watts.
Core2 Duo e6700: 213 watts.
Core2 Duo e6600: 199 watts.

Pentium EE 965: 312 watts.
Pentium D950: 304 watts.
Pentium D840: 344 watts.

Athlon 64 fx 62: 283 watts.
Athlon x2 4400+: 280 watts.
Athlon 64 fx 60: 255 watts.

Still haven't found a list of older hardware listed like this. (We didn't care about power usage in the past?)
ID: 28847 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Hoelder1in
Avatar

Send message
Joined: 30 Sep 05
Posts: 169
Credit: 3,915,947
RAC: 0
Message 28849 - Posted: 3 Oct 2006, 4:50:51 UTC - in response to Message 28847.  
Last modified: 3 Oct 2006, 4:51:06 UTC

Core2 Duo e6700: 213 watts
...just checked the power comsumption of my Core 2 Duo E6700: my watt-meter claims that it uses just 135 Watts (while crunching Rosetta with monitor turned off) - it is certainly not a 'gaming setup' but otherwise fully equipped system (one HD, DVD-Rom, streamer, 'non-gaming' graphics card, 300 W power supply).

Team betterhumans.com - discuss and celebrate the future - hoelder1in.org
ID: 28849 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile River~~
Avatar

Send message
Joined: 15 Dec 05
Posts: 761
Credit: 285,578
RAC: 0
Message 28850 - Posted: 3 Oct 2006, 7:16:56 UTC - in response to Message 28849.  
Last modified: 3 Oct 2006, 7:23:06 UTC

Core2 Duo e6700: 213 watts
...just checked the power comsumption of my Core 2 Duo E6700: my watt-meter claims that it uses just 135 Watts (while crunching Rosetta with monitor turned off)


yes, it is important to check while BOINC is ruinning - using the cpu 100% makes a noticeable difference to the power draw.

Also it is important to check power for a given system, not just to rely on other people's postings. They may have a different mobo, different disks, different number of disks, even a different PSU (they vary in their efficiency, so for the same DC power draw changing the PSU can make a big difference.

EDIT to add: Also you need to use a meter that can add up the energy used over several hours, and average it out. I have found that power use varies on some systems (maybe due to hard drives running up/down, etc?) and is pretty constant on others. The plug in power meter I use cost under £20 from Maplin (not a cheap store) and that will display total enrgy used since start.

On the power costs, I go with Feet1st: I don't cost the power as I only run the full farm in winter and evey kWh burnt goes to heat my lounge, which saves on the other heating costs. It does mean that my RAC drops to a very low level over summer, and in an exceptionally warm Autumn like we are having in the UK at present.

River~~
ID: 28850 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Number crunching : Math for Farmers



©2024 University of Washington
https://www.bakerlab.org