mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2005-01-14, 08:26   #1
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

52910 Posts
Exclamation POWER CONSUMPTION idle versus Prime95

WARNING: HIGH VOLTAGE!

I have been experimenting with a power analyser Model2000MU-UK made by www.prodigit.com

This device is inserted between the mains electricity supply and the device under test, and displays:

Voltage(V)
Current in Amps(A)
Active Power in Watts(W)
Apparent Power in Volt-Amps(VA)
AC Frequency(Hz)
Power Factor(PF)
Energy quantity of consumption(KWhour)
Time elapsed this test.

Voltage and current are RMS (root mean square) values.

There have been guesses elsewhere on this forum as to how much extra electricity a business would consume by joining primenet versus sitting idle.
My intent is not to discourage running Primenet but to make available some factual information on the basis of which to make such a decision.

I'm also interested in how much it costs to run my computers 24x7, and which ones are more efficient than others. By removing components I can measure the effect this has on power consumption.

After lots of experimentation and measurement I will post my findings here.

When reading the results, KWh values are comparable anywhere, but remember that here in the UK our voltage is nominally 240volts at 50Hz.

If you live in the USA or use a different voltage supply the Amps you experience will be different because the voltage is different.

I would be interested in anyone else doing comparative measurements or comments on my results.
Peter Nelson is offline   Reply With Quote
Old 2005-01-14, 09:08   #2
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Arrow Tests on four computers

I have tested:
A)Toshiba Satellite Pro 4280 laptop (pentium III at 497MHz)
B)Compaq Deskpro P733 (pentium III at 730MHz)
C)Self-build Celeron D320 (2.4Ghz overclocked to 2.7 GHz) on Gigabyte 865G board 2x512MB DDR socket 478
D)Self-build Pentium 4 @ 3.0GHz NOT overclocked on Intel 865G board+ATI9600XT graphics 2x512MB DDR socket 478

Calculations assume 24x7 operations. Simply scale down if you only power up during the day etc. Voltage was in the range 238V - 243V but relatively stable. All readings vary a little, and are subject to rounding and reading errors. Decimal places are arbitrary, at my discretion.

Costs are calculated based on 5 pence per unit (KWhour) which is roughly the charge for residential customers in the UK. Substitute your own cost per KWH in your own currency as appropriate. Each configuration was run for up to 8 hours in some cases and consumption scaled up to a day and year for comparison.

It's very cool how the watts, amps etc jump up when mprime or prime95 is invoked.

PENTIUM III@500MHz

System A:0.5333KWH/day=194.8KWH/year without Prime @ 5p=£9.74pa
System A:0.6353KWH/day=232.04KWH/year RUNNING Prime @ 5p=£11.60pa
EXTRA 37.24 KWH/year = EXTRA £1.86/year

For info A idle: 0.18A 21W 43VA
For info A running: 0.3A 38W 75VA

PENTIUM III@733MHz

System B:0.8426KWH/day=307.74KWH/year without Prime @5p=£15.39pa
System B: 1.1623KWH/day=424.54KWH/year RUNNING Prime @5p=£21.23pa
EXTRA 116.8 KWH/year = EXTRA £5.84/year

For info B idle:0.27A 35W 66VA
For info B running: 0.36A 50W 86VA

CELERON D@2.7Ghz

System C:2.09KWH/day=763.49KWH/year without Prime @ 5p=£38.18
System C:3.69KWH/day=1347.77KWH/year RUNNING Prime @5p=£67.39
EXTRA 584.82KWH/year=EXTRA £29.21/year

For info C idle 0.61A 91W 148VA
For info C running 1.07A 155W 258VA

PENTIUM4(NORTHWOOD)@3.0GHz

System D:2.4KWH/day=876KWH/year without Prime @5p=£43.83
System D:4.1143KWH/day=1502.74KWH/year running Prime @5p=£75.14
EXTRA 626.74KWH/year=£31.31/year

For info D idle 0.5A 96W 120VA
For info D running 0.85A 168W 207VA

CONCLUSION

For typical systems in use today, running software for primenet does incur additional cost due to the rise in electricity consumed. Such increases can be quantified. It is worthwhile to compare the energy consumption of different hardware configurations.

I have seen TCO (total cost of ownership) calculated over 3 years.
When making heavy use of the cpu (as in Prime95/Mprime) the cumulative cost of electricity may approach the purchase cost of the components or system.
Peter Nelson is offline   Reply With Quote
Old 2005-01-14, 09:24   #3
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

21116 Posts
Default Power consumption and overclocking

I experimented with the Celeron D 320 system (designed for 2.4GHz)
I usually run it at 2.7GHz (overclocked moderately but it can also be overclocked to 3GHz or more.

For info, System C(CeleronD) runs with cpu temp around 34 degrees C.
System D(Northwood P4) runs with cpu temp around 44 degrees C.
Both are using Intel's air cooling.

In the test above it is recorded:
idle: 0.61A 91W 148VA
running: 1.07A 155W 258VA

My next step was to remove some surplus hardware
Extra cooling fan (leaving the one on the cpu and case)
Four way fan controller
SATA to IDE adapter board

This dropped the power to:
idle: 0.59A 83W 136VA
running: 1.00A 147W 243VA

The above was at 2.7GHz.

I also tried the unclocked 2.4 speed:
idle: 0.56A 83W 136VA
running: 0.92A 138W 227VA

And overclock to 3GHz speed:
idle: 0.6A 87W 144VA
running: 1.07A 158W 261VA

So, as you overclock the system it uses more power.
Peter Nelson is offline   Reply With Quote
Old 2005-01-14, 09:31   #4
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Lightbulb Saving power by removing hard and optical drives

My next optimisation was to remove my hard drive and DVD drive from the Celeron system.

I was able to boot linux using a floppy disk which also contained the prime software.

Now, the idle system consumes:
0.51A 73W 124VA
And when running mprime:
0.91A 136W 222VA

This was at my standard overclock of 2.7GHz

It's also worth noting that if you didn't just leave the system idle but put it in or let it timeout and powerdown into standby mode, it only uses
0.06A 6W 15VA
Removing the power plug would reduce this to zero but neither of these will get any exponents tested.
Peter Nelson is offline   Reply With Quote
Old 2005-01-14, 09:32   #5
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

16916 Posts
Default

Results are as expected.

Another interesting calculation would be the cost per iteration using a 1280K FFT or so.
Dresdenboy is offline   Reply With Quote
Old 2005-01-14, 09:37   #6
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Lightbulb Taking out the VGA and hyperthreading power consumption

I took the same approach with my P4

I removed the 9600XT graphics card and used the onboard Intel extreme graphics 2 in the chipset instead.

On the P4 system I tried measuring consumption with and without Hyperthreading enabled. There was no difference to power in the idle system and minimal increase when running mprime.

This gave:
0.46A 87W 111VA when idle
0.81A 159W 194VA running prime, no hyperthreading
0.82A 163W 201VA running prime, hyperthreading enabled.

I turned hyperthreading back off again the way the system normally is.
Peter Nelson is offline   Reply With Quote
Old 2005-01-14, 09:43   #7
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Lightbulb Removing drives again to reduce power consumption

The P4 system contained two SATA hard drives and a PCI card to support them. There was also a DVD optical drive.

Taking the same approach as with the Celeron D system I removed these components and once again tried running Mprime from a linux boot floppy.

Now the results were:
P4 driveless idle: 0.34A 60W 82VA
P4 driveless running: 0.68A 132W 162VA

Conclusion: a system using network or floppy booting will be more energy efficient for running so is desirable if these components are not required for other uses of the machine.

In comparing the CeleronD and P4 systems, please note the PSUs were both rated at "550W" but differed in manufacturer and quality.

The CeleronD used a QTEC PSU while the P4 used a more expensive (and maybe more efficient) PSU branded EZ-COOL.
Peter Nelson is offline   Reply With Quote
Old 2005-01-14, 09:50   #8
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Post Speed comparison for the above systems

I have given details and power consumption for my P4 and Celeron D systems above.

I here provide iteration speed benchmarks. From these you could work out electricity cost per iteration or electricity cost per exponent.

Time for iteration in ms (FFT size, my P4 left, my CeleronD right)

384K 11.893, 16.227
448K 14.217, 19.759
512K 16.250, 21.899
640K 19.439, 26.783
768K 23.719, 32.290
896K 28.028, 39.116
1024K 31.303, 42.686
1280K 40.896, 56.825
1536K 50.533, 69.411
1792K 59.986, 87.102
2048K 67.633. 97.131

The Celeron D isn't as fast but it's a cheaper system to build and uses less power.
Peter Nelson is offline   Reply With Quote
Old 2005-01-14, 10:27   #9
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Smile Cost of testing an exponent

Dresdenboy,

Thanks for your question about cost per iteration, (which will be miniscule), I think you're really interested in how much electricity bill money to test an exponent.

OK, you will see I have now posted the time per iteration for various FFT sizes on the machines I tested so you can do exactly that calculation.

You were interested in 1280K FFT size, which on SSE2, is used for exponents in the range 20.05 million through 24.93 million.

A day's processing on the Celeron D costs 3.69KWH*5p=18.45pence (feel free to convert to Euros if you want)
In this day, it can test 1520457 iterations at 56.825msec each.

In the same day, my P4 costs 4.1143KWH*5p=20.57pence to run

And it will test 2112676 iterations at 40.896msec each.

Bigger exponents require more iterations you can either multiply up the time per iteration to get time elapsed for a LL test OR you could note how long a test takes (couple of months?) and multiply by the cost of electricity per day.

Two months CeleronD costs £67.39/year divided by 6 = £11.23
Two months Pentium4 (note would do 38% more work in this time) but costs £75.14/6=£12.52

Also note the P4 system cost quite a bit more than the Celeron, maybe £100 extra for the processor? circa £150 vs £50 for the Intel chip excluding other components.

This comparison is for my bare figures before I started stripping out drives and graphics cards.
Peter Nelson is offline   Reply With Quote
Old 2005-01-14, 11:18   #10
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

192 Posts
Default

So the P4 costs 9.74p per 1M iterations, while the C-D costs 12.13p for the same amount of work.

But still - as you've already pointed out - the higher cost for the P4 would surely outweigh these savings.
Dresdenboy is offline   Reply With Quote
Old 2005-01-16, 19:42   #11
patrik
 
patrik's Avatar
 
"Patrik Johansson"
Aug 2002
Uppsala, Sweden

23·53 Posts
Default

I earlier posted some measurements here but since formatting changed when the forums were moved (or upgraded?), I repost it here. I didn't have time to make any further measurements of my newer computers.

All below are P4 Northwoods:
Code:
      DELL Dimension 4550    Tyan Trinity 510	ASUS P4S333
		2399 MHz	1907 MHz	2571 MHz
Off		 10 W		  8 W		 11 W
Idle		 87 W		 76 W		 76 W
ECM Stage 1	135 W		110 W		  -
P-1 Stage 1	145 W		118 W		135 W
P-1 Stage 2	142 W		  -		131 W
LL test		144 W		118 W		135 W
patrik is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
TDP vs. actual power consumption with Prime95 running? ixfd64 Hardware 6 2012-08-31 02:58
power consumption Unregistered Information & Answers 2 2011-01-27 20:48
Now power consumption numbers using Prime95 Dresdenboy Hardware 1 2004-11-23 18:29
Power consumption dans Hardware 7 2004-06-06 15:15
Power consumption optim Hardware 8 2003-12-06 04:13

All times are UTC. The time now is 20:34.

Sun Nov 29 20:34:25 UTC 2020 up 80 days, 17:45, 4 users, load averages: 1.35, 1.21, 1.19

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.