mersenneforum.org  

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

Reply
 
Thread Tools
Old 2016-12-21, 13:58   #1
cybergreg
 
Dec 2016

1010 Posts
Default Unexpected speed difference

I am completely new to all this so please understand
I am running Prime95 on two systems and the "slower" system is running 3 x faster than the "fast" system.

The "slow" system is an HP Z600 workstation running Win 10 pro 64-bit.
Intel Xeon X5570 @ 2.93GHz, 2 processors
RAM 12.0 G 1333MHz fsb
C:\ drive 150Gb SSD
Prime95 is giving approx. 9ms per iteration.

The "fast" system is a DELL T610 server running Win 10 pro 64-bit.
Intel Xeon X5675 @ 3.07 GHz, 2 processors.
RAM 48 G 1333MHz fsb
C:\ drive 1Tb Western Digital Black
Prime95 is giving approx. 28ms per iteration.

Now as I bought the DELL T610 server specifically to run Prime95 this is a bit of a disappointment. The only thing that makes sense from what I can see is that the Western Digital HDD is slugging the system. Is that right? Does Prime95 do a lot of read/writing to the main drive? I would have thought that it would just use the RAM in its calculations - but as I said at the beginning - I know nothing about this stuff.

Can anybody help me out with what's going on here??
cybergreg is offline   Reply With Quote
Old 2016-12-21, 15:59   #2
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

7·13·61 Posts
Default

How many threads per test, and how many separate tests on each system? One copy of P95, or multiple instances?
VBCurtis is offline   Reply With Quote
Old 2016-12-21, 16:03   #3
axn
 
axn's Avatar
 
Jun 2003

34·67 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
How many threads per test, and how many separate tests on each system? One copy of P95, or multiple instances?
Also, what are the numbers that are being tested (specifically, are some of them much smaller/larger than the others)?

Also, are the memory properly configured to run in triple channel mode?
axn is online now   Reply With Quote
Old 2016-12-21, 16:40   #4
cybergreg
 
Dec 2016

2·5 Posts
Default

Thank you for looking at this problem guys.
I am a rank beginner and don't know what I'm doing so can't answer all your questions.
The numbers that are being tested are eight digits long on both machines starting either 43 or 44 - so they aren't wildly different.
There are 3 workers running on the DELL server and 2 running on the HP Z600.
I think there are 2 threads per worker but wouldn't swear to it (I can check if you can tell me where to look).
There is only one instance of Prime95 running on each system.
I haven't got a clue about how the memory is configured to run - isn't that all sorted out by the OS???
cybergreg is offline   Reply With Quote
Old 2016-12-21, 16:49   #5
cybergreg
 
Dec 2016

A16 Posts
Default

I've just seen that there are 4 threads per worker on each system
cybergreg is offline   Reply With Quote
Old 2016-12-21, 16:54   #6
cybergreg
 
Dec 2016

2×5 Posts
Default

My previous post hasn't appeared, but the one I made a few seconds back is already up??
The numbers being tested on both systems are both 8 digits long and start either 44 or 43 so the numbers are not wildly different.
Only 1 instance of Prime95 on each system
3 workers with 4 threads on the DELL server.
2 workers with 4 threads on the HP Z600.
I have no idea about memory configuration - isn't that all handled by the OS??
cybergreg is offline   Reply With Quote
Old 2016-12-21, 17:16   #7
fivemack
(loop (#_fork))
 
fivemack's Avatar
 
Feb 2006
Cambridge, England

11001001101102 Posts
Default

The faster-running system has two chips, each with four CPUs on; Prime95 is running two workers, each using four CPUs, so each worker sits on its own chip.

The slower-running system has two chips, each with six CPUs on; Prime95 is running three workers, each using four CPUs, so one of the workers is having to do masses of inter-chip communication, which is not super-efficient on any system and is particularly bad on that generation of Xeon system.

You need to convince it to run four workers each using three CPUs, or two each using six. There ought to be a file called local.txt or local.ini in the Prime95 directory with lines

Code:
WorkerThreads=3
ThreadsPerTest=4
which you should change to 4/3 or 2/6 (try both, see which goes better)
fivemack is offline   Reply With Quote
Old 2016-12-21, 17:17   #8
Mark Rose
 
Mark Rose's Avatar
 
"/X\(‘-‘)/X\"
Jan 2013

56408 Posts
Default

I would run only a single worker on each and then compare again.
Mark Rose is offline   Reply With Quote
Old 2016-12-21, 17:40   #9
cybergreg
 
Dec 2016

2×5 Posts
Default

Quote:
Originally Posted by fivemack View Post
The faster-running system has two chips, each with four CPUs on; Prime95 is running two workers, each using four CPUs, so each worker sits on its own chip.

The slower-running system has two chips, each with six CPUs on; Prime95 is running three workers, each using four CPUs, so one of the workers is having to do masses of inter-chip communication, which is not super-efficient on any system and is particularly bad on that generation of Xeon system.

You need to convince it to run four workers each using three CPUs, or two each using six. There ought to be a file called local.txt or local.ini in the Prime95 directory with lines

Code:
WorkerThreads=3
ThreadsPerTest=4
which you should change to 4/3 or 2/6 (try both, see which goes better)
Many thanks for your suggestion!
cybergreg is offline   Reply With Quote
Old 2016-12-21, 17:42   #10
cybergreg
 
Dec 2016

2×5 Posts
Default

Quote:
Originally Posted by Mark Rose View Post
I would run only a single worker on each and then compare again.
That too is a good idea - thank you.
cybergreg is offline   Reply With Quote
Old 2016-12-21, 18:56   #11
S485122
 
S485122's Avatar
 
"Jacob"
Sep 2006
Brussels, Belgium

5·7·53 Posts
Default

Before editing the configuration files (prime.txt or local.txt) or the worklist file (worktodo.txt), you must first stop the Prime95 (menu "Test", then "Stop".) You can also completely exit.

You can also change the number of worker windows by using the menu : "Test" "Worker Windows".

If you decrease the number of worker windows you will have too many worker entries in the file worktodo.txt. Stop Prime95, edit the file and remove the line "[Worker #3]". The content of that queue will be attributed to [Worker #2]...

The file undoc.txt contains a lot of information about the content of the configuration files...

Jacob
S485122 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Unexpected exponent expiration Siegmund PrimeNet 6 2017-10-03 03:21
Unexpected Factor apocalypse PrimeNet 6 2015-03-15 19:31
unexpected excrement fivemack Msieve 5 2011-05-09 23:38
Unexpected jackpot fivemack Factoring 6 2010-11-10 19:34
Unexpected ECM factor from Prime95 philmoore Factoring 19 2007-07-10 13:19

All times are UTC. The time now is 14:45.


Thu Dec 1 14:45:37 UTC 2022 up 105 days, 12:14, 1 user, load averages: 1.09, 0.96, 1.01

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, 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.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔