mersenneforum.org  

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

Reply
 
Thread Tools
Old 2009-03-22, 07:56   #1
lidocorc
 
lidocorc's Avatar
 
Nov 2008
Rosenheim, Germany

1816 Posts
Default Overclocking and reliability

Recently I chose new settings for CPU and RAM which made my computer 10% faster. Today the computer finished an exponents's LL-double-check. I looked up the "exponent status" in GIMPS and found the checksum (or watermark?) of my result different from the one of the first run LL-test. It looks suspicious like the increase of speed caused my computer to calculate a wrong result. But there were no round off error messages during the run.

I ask myself: Was the non-satisfying LL result due to overclocking the computer? I could go back to the old CPU and RAM settings and try the same exponent again. This would take 2 weeks and more. Isn't there a cheaper and faster method to find out what the reason was?

lidocorc
lidocorc is offline   Reply With Quote
Old 2009-03-22, 12:57   #2
markr
 
markr's Avatar
 
"Mark"
Feb 2003
Sydney

3·191 Posts
Default

Yes - prime95 has a very good torture test option built in. Just stop your current test first, then you'll find it under the options menu. The readme.txt & stress.txt files have more information. Let it run for a while - stress.txt recommends 6 to 24 hours. Any error means there is a problem.
markr is offline   Reply With Quote
Old 2009-03-22, 13:06   #3
lidocorc
 
lidocorc's Avatar
 
Nov 2008
Rosenheim, Germany

23·3 Posts
Default

Thank you. I'll try that now.
lidocorc is offline   Reply With Quote
Old 2009-03-22, 17:47   #4
Jeff Gilchrist
 
Jeff Gilchrist's Avatar
 
Jun 2003
Ottawa, Canada

7·167 Posts
Default

Another good stress tester that gets CPUS even hotter than Prime95's torture test is the IntelBurnTest. You can get more info and download it from here:
http://www.xtremesystems.org/forums/...d.php?t=197835

It will usually catch stability problems much faster than Prime95 will.
Jeff Gilchrist is offline   Reply With Quote
Old 2009-03-22, 19:42   #5
cheesehead
 
cheesehead's Avatar
 
"Richard B. Woods"
Aug 2002
Wisconsin USA

22·3·641 Posts
Default

Quote:
Originally Posted by Jeff Gilchrist View Post
Another good stress tester that gets CPUS even hotter than Prime95's torture test is the IntelBurnTest. You can get more info and download it from here:
http://www.xtremesystems.org/forums/...d.php?t=197835
I hadn't heard of IntelBurnTest before. Interesting.

For those of you familiar with LINPACK: It's based on Linpack by Intel(R). IntelBurnTest is "an easier to use linpack".

For those of you not familiar, "LINPACK is a collection of Fortran subroutines that analyze and solve linear equations and linear least-squares problems" (http://www.netlib.org/linpack/). The LINPACK Benchmark is used to rank the performances of the Top 500 supercomputers (http://www.top500.org/project/linpack). Here's a FAQ: http://www.netlib.org/utk/people/Jac...q-linpack.html

I've looked at only the first couple and last pages of the 32-page thread so far and notice:

1. There's a tendency for commenters to compare IntelBurnTest to Prime95's small-FFT test. That's not right; the comparison should be to Prime95's Blend test (some posters get it right), I think. IntelBurnTest is designed to use maximum RAM, I read. (OTOH, one poster claims IBT doesn't stress RAM at all! There may be confusion, or maybe IBT reads/writes RAM less frequently than P95 while using a large memory range.)

2. IntelBurnTest gets your CPU hot (if you let it). Someone gave a guideline of 20C hotter than Prime95. Someone else reported 93C. (I think some of those folks are going to damage CPUs that were perfectly fine before they ran IntelBurnTest with thermal cutoffs raised or disabled. It isn't IntelWarmTest. Do you really need to know whether your FPU glitches at 93C if it never glitched before at 70C?)

Quote:
It will usually catch stability problems much faster than Prime95 will.
One poster had a system pass IntelBurnTest just fine, but fail quickly with prime95 small-FFT.

Last fiddled with by cheesehead on 2009-03-22 at 20:14
cheesehead is offline   Reply With Quote
Old 2009-03-22, 22:37   #6
Cruelty
 
Cruelty's Avatar
 
May 2005

2×809 Posts
Default

I've run this software. It makes my CPU 6-7 degrees hotter than Prime95 and only 3-4 hotter than running 4 instances of LLR. Anyways, it might be used as a quicker way to check CPU stability.
Cruelty is offline   Reply With Quote
Old 2009-03-23, 13:48   #7
Jeff Gilchrist
 
Jeff Gilchrist's Avatar
 
Jun 2003
Ottawa, Canada

7×167 Posts
Default

Quote:
Originally Posted by cheesehead View Post
One poster had a system pass IntelBurnTest just fine, but fail quickly with prime95 small-FFT.
I think I found IBT raised the temp of my CPU cores around 7-10C hotter than Prime95 did. It depends on how many tests the person used for IBT as well.

I would recommend running both, but in my case IBT would fail within 10-20 minutes while Prime95 might take 3-4 hours for the same failure to be detected.

But I did both. Once I thought I found a good stable setting for running 24/7, I did 8 hours of IBT testing, then about 12-24 hours of Prime95 torture test, then I did Prime95 double checks (4, 1 for each core on my quad) from the server and verified my resides to make sure they matched. That gave me about 1 week of 24/7 work with the double checks so I was satisfied it was stable after that.

Jeff.
Jeff Gilchrist is offline   Reply With Quote
Old 2009-03-24, 07:18   #8
lfm
 
lfm's Avatar
 
Jul 2006
Calgary

1A916 Posts
Default

Quote:
Originally Posted by Jeff Gilchrist View Post
Another good stress tester that gets CPUS even hotter than Prime95's torture test is the IntelBurnTest. You can get more info and download it from here:
http://www.xtremesystems.org/forums/...d.php?t=197835

It will usually catch stability problems much faster than Prime95 will.
It is a good test but Prime95 also needs your memory to work at extreme performance levels and the CPU burn tests do not test memory.
lfm is offline   Reply With Quote
Old 2009-03-24, 12:38   #9
Jeff Gilchrist
 
Jeff Gilchrist's Avatar
 
Jun 2003
Ottawa, Canada

7×167 Posts
Default

Quote:
Originally Posted by lfm View Post
It is a good test but Prime95 also needs your memory to work at extreme performance levels and the CPU burn tests do not test memory.
Which is why I recommend doing *both*. I would disagree that it doesn't test memory though. IBT uses up something like 80% of your RAM for its calculations and if something in RAM gets corrupted the results will not match so you will get errors from faulty RAM. It might not be designed specifically to test memory, but bad memory will cause it to fail.
Jeff Gilchrist is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Reliability and Confidence of CPUs? AlexJohn73 Information & Answers 4 2015-11-22 09:26
what are Reliability and Confidence? dragonbud20 Information & Answers 10 2015-10-21 03:26
nvidia card reliability Roy_Sirl GPU Computing 14 2012-07-23 13:51
Reliability and confidence level lidocorc Information & Answers 6 2009-08-11 04:04
NewPGen reliability Cruelty Riesel Prime Search 3 2006-02-15 05:15

All times are UTC. The time now is 08:17.

Thu Jun 4 08:17:22 UTC 2020 up 71 days, 5:50, 0 users, load averages: 1.58, 1.45, 1.39

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.