mersenneforum.org  

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

Reply
 
Thread Tools
Old 2019-12-30, 03:13   #1
guilhermeibc
 
Dec 2019

116 Posts
Default Prime95 throughput vs Power Consumption

Hi everyone,


First of all, because this is my first post in this forum, please let me know if this thread should be moved somewhere else.


In a nutshell, if you want to skip reading through, check my graph below summarizing all of my findings. I've also attached this image and the Excel spreadsheet to this thread.



https://imgur.com/a/bjbjAtm


I have quite a few different computer systems at my disposal and I decided to run the Prime95 benchmark on all of them, while measuring their power consumption using the widely known " Kill-a-watt-meter" (http://www.p3international.com/products/p4400.html)


After reading several posts in this forum, I consider myself a beginner when it comes to advanced approaches, fine tuning, etc. Nevertheless, I'd like also to say that I've been aware and an occasional user of the GIMPS project since 2004.


Test environment and considerations:
  • All of the benchmark tests were run using a fresh, USB bootable version of the Porteus Linux. Then, Prime95 was downloaded and run.
  • With the intent to achieve the best performance possible. the Linux system was run in text mode only.
  • In every computer system, Prime95 - Linux: 64-bit (29.8b6) version was downloaded and uncompressed.
  • All benchmark tests were run with their default settings as "desired by mPrime". Therefore, no benchmark test parameters we modified, although their default values would vary slightly across different computers.
The computer systems tested:
  • Lenovo laptop W530 - Intel(R) Core(TM) i7-3840QM CPU @ 2.80GHz
  • Dell laptop XPS L502x - Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz
  • Dell laptop Vostro 3300 - Intel(R) Core(TM) i3 CPU M 370 @ 2.40GHz
  • Dell PowerEdge 2950 - Dual Socket Intel(R) Xeon(R) CPU 5130 @ 2.00GHz
  • Dell PowerEdge 1950 - Dual Socket Intel(R) Xeon(R) CPU E5205 @ 1.86GHz
  • Dell PowerEdge T110 II - Intel(R) Xeon(R) CPU E3-1220 V2 @ 3.10GHz
  • Sun SunFire X2100 M2 Server - Dual-Core AMD Opteron(tm) Processor 1210
  • HP Proliant DL320 G5 Server - Intel(R) Xeon(R) CPU 3050 @ 2.13GHz
  • IBM xSeries 306 server - Intel(R) Pentium 4 @ 3GHz
Final notes on the tests performed:
  • The data for the tests on the Vostro 3300 Laptop and the IBM xSeries server is not 100% complete. I've been trying to find time to fill those in my spreadsheet for a while now. Therefore, I decided the post the info I currently have instead of postponing this for too long. Still, I really want to conclude them to see all the results of the oldest player in this lineup, the IBM server with its Pentium 4 processor.
  • The Dell 2950 PowerEdge ran with both redundant power supplies.
  • The only server that had Hard Drives installed was the HP Proliant DL320 G5. The other servers didn't, which might help bringing the power consumption down a little bit.
My conclusions and thoughts and potential questions:
  • For every computer, I noticed that I was able to draw more current/power while running actual Double Check LL tests, as opposed to running torture tests on each computer. I did try all options for the torture tests to see if any would be able to beat reall LL tests. Any reasons for that?
  • When running mPrime on the servers, I did not notice any of their fans ramping up to high speeds when running Double Check LL tests, torture tests or benchmark tests. They all seemed to run at the same speed as if the servers where idling. Personally, that was something interesting to observe.
  • As expected, based on all of the throughput results, the best results were running 1 worker per physical CPU socket and having the # of helpers (or workers as seen in the benchmark logs) equal to the # of existing cores per CPU.
  • Also as expected, newer hardware is more efficient, which to me relates with the CPU's clock's and their nanometer lithography process (forgive me for wrong use of terminology).
  • Performing this test led me to conclude that, even when having all these computers hanging around, there is a lot of power consumption to be considered if I wanted to have them all running 24/7 just for the GIMPS purpose. I would definitely do that If I had free power! Old hardware is cheap for a reason.
  • I do have plans run this same tests with newer computers, Raspberry Pi's and/or other SBC's.
Thanks for reading and let me know if you have any questions.
Attached Thumbnails
Click image for larger version

Name:	Comparison.png
Views:	113
Size:	85.6 KB
ID:	21509  
Attached Files
File Type: pdf PowerTest.pdf (195.5 KB, 114 views)
guilhermeibc 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 idle versus Prime95 Peter Nelson Hardware 10 2005-01-16 19:42
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 06:27.

Sun Mar 7 06:27:45 UTC 2021 up 94 days, 2:39, 0 users, load averages: 2.39, 2.57, 2.21

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.