mersenneforum.org  

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

Reply
 
Thread Tools
Old 2020-02-15, 15:21   #45
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

66268 Posts
Default

For some extensive benchmarking tabulations of various cpu models in prime95 with various cores/worker and fft lengths, see https://www.mersenneforum.org/showpo...19&postcount=5 and https://www.mersenneforum.org/showpo...18&postcount=4. Ranging from dual-core Core2 duo e8200 to dual-12-core e5-2697 Xeon.
kriesel is offline   Reply With Quote
Old 2020-02-19, 10:51   #46
jas
 
"Simon Josefsson"
Jan 2020
Stockholm

3×5 Posts
Default

Quote:
Originally Posted by kriesel View Post
For some extensive benchmarking tabulations of various cpu models in prime95 with various cores/worker and fft lengths, see https://www.mersenneforum.org/showpo...19&postcount=5 and https://www.mersenneforum.org/showpo...18&postcount=4. Ranging from dual-core Core2 duo e8200 to dual-12-core e5-2697 Xeon.

Thank you for the links!



This seems to confirm my findings (or do you disagree?):


- make sure all memory channels are populated (significant performance differences!)


- that HT is not useful


- use all cores


- adding more workers beyond 4 does not lead to significant better performance and that hardware failures etc become more problematic.



Before reading your graphs, my preference was to use 2 workers with all cores, no HT, but after reading it I think I may prefer 3 or 4 workers instead. I'm doing benchmarks on a 2x2640v2 system now and it suggests a 4 worker setup too.



/Simon
jas is offline   Reply With Quote
Old 2020-02-19, 14:36   #47
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2×37×47 Posts
Default

Quote:
Originally Posted by jas View Post
Thank you for the links!

This seems to confirm my findings (or do you disagree?):

- make sure all memory channels are populated (significant performance differences!)

- that HT is not useful

- use all cores

- adding more workers beyond 4 does not lead to significant better performance and that hardware failures etc become more problematic.

Before reading your graphs, my preference was to use 2 workers with all cores, no HT, but after reading it I think I may prefer 3 or 4 workers instead. I'm doing benchmarks on a 2x2640v2 system now and it suggests a 4 worker setup too.

/Simon
Mostly agree with the preceding. Would modify as follows.

HT use in prime95 is usually not useful. Benchmark it anyway.

HT is VERY useful for keeping other tasks from slowing down prime95; make sure it is enabled in the BIOS and OS. On one system I'm running a single core Mlucas in parallel with all-cores prime95, and see no slowdown from that.

Use at least one worker per chip package. Communication between chip packages is too slow for production use. Only if confirming a reported prime is there any point to sacrificing a lot of throughput to reduce latency. And in some cases using multiple chip packages will INCREASE latency.

Optimal configuration is fft length dependent, system configuration dependent, and often prime95 version dependent, so benchmark again and reassess # of workers etc when changing fft length or system configuration or prime95 version.

The above was for prime95's built in benchmarking. Confirm with actual runs. Question whether P-1 optimal is the same as LL/PRP optimal. Or Mlucas vs prime95/mprime.

Last fiddled with by kriesel on 2020-02-19 at 14:42
kriesel is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
access to windows machine files from ubuntu machine wildrabbitt Hardware 1 2019-02-24 18:34
prime95 settings evanh Software 3 2017-12-04 15:18
CPU settings and other newbie Qs Z3noN Information & Answers 2 2017-01-11 05:58
Memory Settings Fred Software 5 2016-05-03 00:51
settings for cudaLucas fairsky GPU Computing 11 2013-11-03 02:08

All times are UTC. The time now is 08:20.

Tue Apr 7 08:20:48 UTC 2020 up 13 days, 5:53, 2 users, load averages: 1.76, 1.94, 1.93

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.