mersenneforum.org  

Go Back   mersenneforum.org > Prime Search Projects > Riesel Prime Search

Reply
 
Thread Tools
Old 2012-08-17, 22:56   #1
pinhodecarlos
 
pinhodecarlos's Avatar
 
"Carlos Pinho"
Oct 2011
Milton Keynes, UK

23×571 Posts
Default Per-Watt Efficiency

shanecruise, what processor (Intel, AMD, frequency, etc) are you using? Thank you in advance.
pinhodecarlos is offline   Reply With Quote
Old 2012-08-18, 02:06   #2
shanecruise
 
shanecruise's Avatar
 
Mar 2012
Hyderabad, India

3316 Posts
Default

Hi
For this set of primes
I'm running 4 threads on Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz
2 threads on Intel(R) Core(TM) i3-2120 CPU @ 3.30GHz
1 thread on Gtx 580

2829.764 sec. is GPU
2629.234 sec. is when single thread is run on i7-2600
3210.944 sec. is when 4 threads run on i7-260 (goes around 5000 when i use 6 cores)
4073.883 sec. when 2 threads run on i3



Quote:
Originally Posted by pinhodecarlos View Post
shanecruise, what processor (Intel, AMD, frequency, etc) are you using? Thank you in advance.
shanecruise is offline   Reply With Quote
Old 2012-08-18, 09:15   #3
pinhodecarlos
 
pinhodecarlos's Avatar
 
"Carlos Pinho"
Oct 2011
Milton Keynes, UK

23·571 Posts
Default

Running GPU is still a waste of energy on LLR, thank you.
pinhodecarlos is offline   Reply With Quote
Old 2012-08-19, 06:02   #4
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

47·89 Posts
Default

Quote:
Originally Posted by pinhodecarlos View Post
Running GPU is still a waste of energy on LLR, thank you.
If the CUDA-LLR becomes more efficient per-watt than CPU, should we stop running it on CPUs because that is now a waste of energy? Why does this matter?
VBCurtis is offline   Reply With Quote
Old 2012-08-19, 06:08   #5
pinhodecarlos
 
pinhodecarlos's Avatar
 
"Carlos Pinho"
Oct 2011
Milton Keynes, UK

23·571 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
If the CUDA-LLR becomes more efficient per-watt than CPU, should we stop running it on CPUs because that is now a waste of energy? Why does this matter?
The same way you try to save water or when you take a bath you waste like 100 L of water? I work in the energy efficiency field and this is one of my concerns. I feel sorry if it isn't one of yours, it should be of all.
At the end it is your money you waste on electricity but do you know how it was produced? Do you know how much CO2 was emitted? I'm thinking of the world for the next generation...
Finally, I think you should take a visit on the Department of Energy website of your country, USA.

Last fiddled with by pinhodecarlos on 2012-08-19 at 06:15
pinhodecarlos is offline   Reply With Quote
Old 2012-08-20, 18:17   #6
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

47×89 Posts
Default

Carlos-
You did not answer my question about not running on CPUs when CUDA-LLR becomes more efficient. The rest of your arguments are much stronger against the entire prime-testing hobby than they are against using a GPU specifically. Since I'm not going to stop primality testing, and am fully aware of the cost of electricity, I still fail to see the strength of your position that GPUs are a worse idea than any other primality testing, particularly when my GPU is *more* efficient than a P4.

I don't follow your analogy about taking baths, unless you're comparing a bath to a shower. In the case of prime searching, we have thousands of numbers waiting in line for a shower, and I have an unused bathtub on my GPU- one which hardly uses more water than a shower.

There are very few people on this forum who have work-per-watt as their highest priority. if you do, you should be underclocking your CPUs, as the voltage cut improves work-per-watt. But then, you trade off the electricity savings vs the extra cost of the nice fast chip you're not taking full advantage of. For the rest of us, we'll take the maximum work rate we can get at a reasonable cost.
-Curtis
VBCurtis is offline   Reply With Quote
Old 2012-08-20, 18:29   #7
pinhodecarlos
 
pinhodecarlos's Avatar
 
"Carlos Pinho"
Oct 2011
Milton Keynes, UK

23×571 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
If the CUDA-LLR becomes more efficient per-watt than CPU, should we stop running it on CPUs because that is now a waste of energy? Why does this matter?
First question: Yes.
Second question: For you it doesn't matter but for me it does. It's the principal of energy saving or producing more with less energy, that's what we call energy efficiency.

Anyway, I am not criticizing you, you can do whatever you want, it's your money not mine. I just want people to understand at the end of the month when they look at the electricity bill if it was an advantage to run llr on a GPU or on a CPU. You can compare for two different months, one only running a GPU and one running a CPU, taking in consideration the candidates tested. The ratio Watt per candidate is still better for a CPU.

Edit: Now I am trying to understand which CPU is more efficient of these three: i7-3770K vs i7-2600K vs i7-3930K because I want to help on testing some k=5 candidates. Everything points to the i7-3770K, better ROI, better Watt/candidate.

Last fiddled with by pinhodecarlos on 2012-08-20 at 18:43
pinhodecarlos is offline   Reply With Quote
Old 2012-08-21, 06:20   #8
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

2·3·1,423 Posts
Default

Quote:
Originally Posted by pinhodecarlos View Post
Edit: Now I am trying to understand which CPU is more efficient of these three: i7-3770K vs i7-2600K vs i7-3930K because I want to help on testing some k=5 candidates. Everything points to the i7-3770K, better ROI, better Watt/candidate.
I can tell you exactly to few decimals for i7-2600k against one (or two or three) gtx580, just let me find a power meter somewhere in this office (which we should have, we produce some electronic stuff here, and I would need to ask for permission to take it home for a while, and I would plug it on the wall where my wheelbarrow is plugged. The wheelbarrow is the one with i7-2600k). For the other two CPU's i can't help, I wasn't so lucky to hold any of them in my fingers yet.
LaurV is offline   Reply With Quote
Old 2012-08-21, 09:14   #9
pinhodecarlos
 
pinhodecarlos's Avatar
 
"Carlos Pinho"
Oct 2011
Milton Keynes, UK

107308 Posts
Default

Quote:
Originally Posted by LaurV View Post
I can tell you exactly to few decimals for i7-2600k against one (or two or three) gtx580, just let me find a power meter somewhere in this office (which we should have, we produce some electronic stuff here, and I would need to ask for permission to take it home for a while, and I would plug it on the wall where my wheelbarrow is plugged. The wheelbarrow is the one with i7-2600k). For the other two CPU's i can't help, I wasn't so lucky to hold any of them in my fingers yet.
I would suggest the following test in case you have only one gtx580 installed on the i7-2600K machine:

1st: disable HT, measure power with machine idle (CPU idle and GPU idle)
2nd: disable HT, measure power output with 4 instances of LLR for CPU with one of the files from the k=5 reservation. Take note on the time done per candidate.
3rd: disable HT, measure power output with only the GPU running one instance of LLR. Take note on how much backup CPU (%) is used while the GPU is running and also take note on the time done per candidate.

Software can be found here: http://www.bc-team.org/downloads.php?cat=7

In case you have the three cards in the machine put 2 into idle, and follow the steps behind. The final output power measured will contain the power of the two idle cards and it will be ~constant for the three tests.

Last fiddled with by pinhodecarlos on 2012-08-21 at 09:17
pinhodecarlos is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Efficiency in % of your computers 3mg4 PrimeNet 8 2016-03-23 13:12
VIA C3 efficiency ET_ Hardware 4 2007-03-27 21:29
Multiple threads efficiency... Xyzzy Linux 1 2004-12-31 21:03
Energy efficiency for LL markhl Hardware 5 2004-02-04 13:33
Trial-factoring efficiency ThomRuley Lone Mersenne Hunters 4 2003-06-07 04:40

All times are UTC. The time now is 21:27.

Sat May 30 21:27:44 UTC 2020 up 66 days, 19 hrs, 1 user, load averages: 1.28, 1.21, 1.31

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.