mersenneforum.org  

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

Reply
 
Thread Tools
Old 2020-02-16, 16:23   #1
docustompc
 
Feb 2020

23 Posts
Default so many questions!!

so..
I really apologise this stuff has probably been asked before..

Firstly I have many machines (Mine the kids and computers fitted on site)
Some Intel some amd, Some powerhouse some not...

I am trying to work out the optimum setting for all of them

I really like the idea of testing numbers that are yet to be tested rather than double checking other numbers

So.. it seems largely that intel is better than AMD core for core

Q1
I have a i7-9700 16gb ram as my main desktop machine and at 2048, 4096 and 8192k fft 1 worker 8 threads is quicker than 2 workers 4 cores.. and 8 workers 1 core each

Is that normal, is this caused by ram bottleneck?
16gb 2666 mhz single rank ram 4 x 4gb sticks set cl14

Q2
I have a few machines that were running 1 worker 4 threads (amd 3400g, 2200g, intel i5-3470k)

And after benchmarking at 2048,4096, 8192k that 4 workers a core each was slightly better than 1 worker 4cores

However even though they are selected to do LL First Tests, the extra new 3 cores are now doing P-1 Stage 1? why is that?

Q3
what is the most efficient way to do this excluding hardware cost i.e. just power costs
I don't mind buying hardware but the ongoing electricity cost is significant!
I wouldn't mind building a few efficient machines for this specifically.. My desktop and the kids PC's all have power hungry CPU's and GPU's and the kids have PSU's that are not anything rated :-) so they use a lot of electric
Quick acid tests with my watt meters tells me that my desktop pc alone runs at 250w from the wall when running prime this will cost approx £300 a year just for my machine to be on 24/7 - time I would be using it probably 5h a day still over £200 a year (electric is expensive in the UK)

The kids 4 machines plus my server draw around 500w..

Has anyone tested any efficient hardware?

I have more questions but that will do for now!!

I must say it is quite exciting love a bit of math problem solving and computer hardware!! so these 2 go together well !!
docustompc is offline   Reply With Quote
Old 2020-02-16, 18:56   #2
docustompc
 
Feb 2020

23 Posts
Default

'1 worker 8 threads is quicker than 2 workers 4 cores.. and 8 workers 1 core each'

Should have read 1 worker 8 cores not threads!!
docustompc is offline   Reply With Quote
Old 2020-02-16, 21:21   #3
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

7·491 Posts
Default

Welcome.
I'll leave it to others that have similar hardware, to comment on whether yours are doing as expected.
There are extensive benchmark summaries for prime95 at

https://www.mersenneforum.org/showpo...18&postcount=4
https://www.mersenneforum.org/showpo...19&postcount=5
and a little more to come.

It's well established that on systems with multichannel access to ram having all the channels populated helps performance.

Please consider switching from LL first tests to PRP first tests. PRP with GEC is MUCH better at detecting errors, about as fast as LL, and avoids the project-wide 2% overhead of having to run triple checks (and 4th, etc.) when there's an LL test mismatch. If PRP returns probably prime, and that is verified, you still get discovery credit.

Do some of your systems have gpus? Gpu computing in GIMPS is a thing. You may be leaving the majority of your prime-finding throughput idle by only using the cpu side with prime95/mprime or Mlucas. See http://www.mersenneforum.org/showpos...91&postcount=2

There's also cloud computing, both cpu and gpu, some of which is free.
https://www.mersenneforum.org/showpo...06&postcount=1

Regarding costs, there's https://www.mersenneforum.org/showpo...8&postcount=20. Given your electrical costs, you may want to look into down-clocking at the link near the bottom of that post. The economics change noticeably depending on whether there's space heating benefit, cooling requirement, or neither.

There's a lot more info indexed at https://mersenneforum.org/showthread.php?t=24607
kriesel is offline   Reply With Quote
Old 2020-02-17, 06:04   #4
kladner
 
kladner's Avatar
 
"Kieren, ktony"
Jul 2011

3×19×167 Posts
Default

kladner is offline   Reply With Quote
Old 2020-02-17, 11:47   #5
docustompc
 
Feb 2020

816 Posts
Default

Quote:
Originally Posted by kriesel View Post
Welcome.
I'll leave it to others that have similar hardware, to comment on whether yours are doing as expected.
There are extensive benchmark summaries for prime95 at

https://www.mersenneforum.org/showpo...18&postcount=4
https://www.mersenneforum.org/showpo...19&postcount=5
and a little more to come.

It's well established that on systems with multichannel access to ram having all the channels populated helps performance.

Please consider switching from LL first tests to PRP first tests. PRP with GEC is MUCH better at detecting errors, about as fast as LL, and avoids the project-wide 2% overhead of having to run triple checks (and 4th, etc.) when there's an LL test mismatch. If PRP returns probably prime, and that is verified, you still get discovery credit.

Do some of your systems have gpus? Gpu computing in GIMPS is a thing. You may be leaving the majority of your prime-finding throughput idle by only using the cpu side with prime95/mprime or Mlucas. See http://www.mersenneforum.org/showpos...91&postcount=2

There's also cloud computing, both cpu and gpu, some of which is free.
https://www.mersenneforum.org/showpo...06&postcount=1

Regarding costs, there's https://www.mersenneforum.org/showpo...8&postcount=20. Given your electrical costs, you may want to look into down-clocking at the link near the bottom of that post. The economics change noticeably depending on whether there's space heating benefit, cooling requirement, or neither.

There's a lot more info indexed at https://mersenneforum.org/showthread.php?t=24607
That is very very helpful Thanks!!

Will look into this PRP

My home server is currently an a320 mobo with 2200g - may consider upgrading this as it's always on anyway 3950x may be overkill on this board though haha!
docustompc is offline   Reply With Quote
Old 2020-02-21, 00:49   #6
docustompc
 
Feb 2020

23 Posts
Default

Soo..

Really interested in doing some work with my GPU's

Whilst I'm really confident with windows and hardware..

Building and compiling software to run workers on my gpu's looks scary!!

from what i can work out.. (although this info may be out of date..)

TF is very reliable and efficient on a gpu? perhaps not so much LL?

I have no issue what the gpu's do but would like them to do 'something' as the machines are on anyway..

I need more help with how to set this up I have a variety of low, medium and high performance cards from amd and nv

I have trailed through the guides and even started to install the cuda development package for cuda lucas but it said it wanted to reinstall my graphics driver which was a no no for me as the driver I run is flawless -

I feel the gpu software is not nearly as user GUI friendly like prime95...

I need a walkthrough!!

If this needs to go in another thread feel free to move it!!

Thank for any assistance!
docustompc is offline   Reply With Quote
Old 2020-02-21, 01:39   #7
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

76358 Posts
Default

Your walkthrough depends entirely on what GPU chip-brand you have. Nvidia/CUDA and AMD/openCL aren't even the same programs, let alone the same instructions on install and use.

If you're going to ask for help, you should start with telling us what you have. What GPU? What operating system?
VBCurtis is offline   Reply With Quote
Old 2020-02-21, 02:39   #8
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

7×491 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
If you're going to ask for help, you should start with telling us what you have. What GPU? What operating system?
"I have a variety of low, medium and high performance cards from amd and nv "

OP should start at the top of https://mersenneforum.org/showthread.php?t=24607 and start reading. Read at least the posts behind the first dozen reference links for background and the occasional data-rich attachment.

Last fiddled with by kriesel on 2020-02-21 at 02:40
kriesel is offline   Reply With Quote
Old 2020-02-21, 13:51   #9
docustompc
 
Feb 2020

23 Posts
Default

Great stuff

SO..

I have mfaktc running seemingly rather well on a gtx 970 and a gtx 1050

at around 530ghz/day and 220ghz/day respectively
It seems mfakto for amd is not complied and built? i have trailed through the forums and couldn't find a compiled / built version? I'll leave that part to the experts I think

next question is tf the best way to utilise my gpu's for the project as a whole or are there other tasks that are more important I could use them for?

Sorry if my terminology is wrong and this info is out there

Some of the threads for mfakto are over 100 pages long!!

I have looked

Hopefully will be submitting first manual results later today!!

Best Regards
Thankyou for your continued help!!
docustompc is offline   Reply With Quote
Old 2020-02-21, 14:34   #10
docustompc
 
Feb 2020

23 Posts
Default

As mentioned I have a variety of cards
Some AMD some NV.. however all running windows 10 64bit latest builds [always updated]
I have mfaktc running and working well !! be good to get some work on the AMD cards
docustompc is offline   Reply With Quote
Old 2020-02-21, 14:59   #11
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

171268 Posts
Default

Since you are windows, use Misfit
https://www.mersenneforum.org/misfit/
It will take care of getting your work and sending in the results. And if you want to track your results through GPU72 you can do that with Misfit too.
Uncwilly is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Many (gmp-)ecm questions... WraithX GMP-ECM 26 2014-04-03 15:54
GPU LL Questions Primeinator GPU Computing 40 2013-10-20 06:20
Some questions... OmbooHankvald PSearch 3 2005-09-17 19:29
5 questions OmbooHankvald Factoring 6 2005-08-28 19:31
Questions OmbooHankvald Prime Sierpinski Project 2 2005-08-01 20:18

All times are UTC. The time now is 00:24.

Thu Apr 2 00:24:49 UTC 2020 up 7 days, 21:57, 3 users, load averages: 0.97, 1.27, 1.29

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.