Energy efficiency for LL

[markhl]

What is the most energy-efficient PC for LL?
I.e. what system will use the least energy to LL-test an exponent?
We have touched on related topics in the past, and I guess that farm builders have a good handle on this.
I'll use kWh because that is how electricity is billed.

E.g. my PIII 866 MHz takes 0.156 s per iteration when the FFT size is 640K.
Let's say it takes 50 W, so it'll take 20 hours to consume 1 kWh.
In those 20 h it'll do 20 * 3600 / 0.156 = 462000 iterations.
That is a figure of merit: 462000 iterations / kWh.
Beat that :-)

1: Choose the OS: Windows runs fast Prime95, Linux may allow more efficiency but runs slower compiled C code.
2: What CPU is best? Prime95 is well-optimized for P4, but the P4 uses a lot more energy.
3: Save energy on drives by going from hard disk to USB drive or CD or RAM drive?
4: Save on cooling, fans, etc.

[geoff]

Originally Posted by markhl

1: Choose the OS: Windows runs fast Prime95, Linux may allow more efficiency but runs slower compiled C code.

Sorry, I can't let that go uncorrected :-)
mprime on x86-based Linux uses the same assembly code that Prime95 on x86-based Windows uses.
Non-x86 Linux has to use one of the other programs instead of mprime, as would Windows running on non-x86 hardware.

[Angular]

Originally Posted by markhl

What is the most energy-efficient PC for LL?
I.e. what system will use the least energy to LL-test an exponent?
We have touched on related topics in the past, and I guess that farm builders have a good handle on this.
I'll use kWh because that is how electricity is billed.

E.g. my PIII 866 MHz takes 0.156 s per iteration when the FFT size is 640K.
Let's say it takes 50 W, so it'll take 20 hours to consume 1 kWh.
In those 20 h it'll do 20 * 3600 / 0.156 = 462000 iterations.
That is a figure of merit: 462000 iterations / kWh.
Beat that :-)

1: Choose the OS: Windows runs fast Prime95, Linux may allow more efficiency but runs slower compiled C code.
2: What CPU is best? Prime95 is well-optimized for P4, but the P4 uses a lot more energy.
3: Save energy on drives by going from hard disk to USB drive or CD or RAM drive?
4: Save on cooling, fans, etc.

How did you measure the power consumption of your computer or are you just looking for the power consumption of the processor?

Consider the 3.1 GHz P4 from the Benchmark thread: ntel(R) Pentium(R) 4 CPU 2.60GHz
CPU speed: 3133.24 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: 8 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 128 bytes
TLBS: 64
Prime95 version 23.4, RdtscTiming=1
Best time for 384K FFT length: 11.437 ms.
Best time for 448K FFT length: 13.554 ms.
Best time for 512K FFT length: 15.425 ms.
Best time for 640K FFT length: 18.535 ms.
Best time for 768K FFT length: 22.552 ms.
Best time for 896K FFT length: 26.662 ms.
Best time for 1024K FFT length: 29.951 ms.
Best time for 1280K FFT length: 39.310 ms.
Best time for 1536K FFT length: 48.347 ms.
Best time for 1792K FFT length: 57.434 ms.
Best time for 2048K FFT length: 64.950 ms.

System draws 70W idle and about 140W with prime95 torture-test - that is about 9.1Ws/2048FFT.

http://www.mersenneforum.org/showthr...9&page=2&pp=25


It does 640K FFTs in 19 ms and consumes 140 W/hr => 7.14 hrs / kWh.
7.14 hrs * 3600 / 0.019 ~=> 1.36 million iterations / kWh

The 140 Watts seams a bit low, but even with a 50% increase it still wins against that P3. I wonder if the athlon is competative. It looks like even the hot P4 is worth the money in work/$.

[markhl]

Originally Posted by Angular

How did you measure the power consumption of your computer

I did not measure; that was just an estimate from others who have put a power meter on their systems. I thought that a good P4 might well beat a P3, but thanks for sharing your measurements.

[Angular]

Originally Posted by markhl

I did not measure; that was just an estimate from others who have put a power meter on their systems. I thought that a good P4 might well beat a P3, but thanks for sharing your measurements.

Your welcome, but those are not my measurements. I found them in the Benchmark thread. I also found someone using a reportedly inaccurate DMM measurement method that said about ~175 Watts for a P4 ~ 3 GHz. Thus my concern about the accuracy of the numbers I used.

Since the Kill a Watt meters are only around $30, I may order one to check my P4 and Athlon systems.

[TauCeti]

Originally Posted by Angular

Your welcome, but those are not my measurements. I found them in the Benchmark thread. I also found someone using a reportedly inaccurate DMM measurement method that said about ~175 Watts for a P4 ~ 3 GHz. Thus my concern about the accuracy of the numbers I used.
Since the Kill a Watt meters are only around $30, I may order one to check my P4 and Athlon systems.

I was my measurement with an acurate power-meter measuring only the real power (and not the reactive power).

If i had used a non-PFC PSU i may have even gotten lower into the 130W range.

More on PFC in my infamous PFC rant and here