mersenneforum.org  

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

Reply
 
Thread Tools
Old 2005-01-16, 00:24   #1
madbullet
 
Jan 2005

23 Posts
Default Athlon prime, strange results

hi, i posted in the hardware section that i was having failings in all tests of prim. now ive tested the athlon version i have some strange results

both Blend and In-place large FFT's fail instantly, how ever the Small FFT's doesn't.

could someone with some knowledge tell me what could be the problem, someone did reply in the previous post o say that it could be a bad CPU because FFT's stress the CPU more (or some words to that affect)

thanks
madbullet is offline   Reply With Quote
Old 2005-01-16, 03:04   #2
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

32×823 Posts
Default

98% likely that it is memory. Small FFTs run out of the L2 cache, blend and in-place must access main memory.

Try memtst86. Try slower RAS, CAS, etc. timings. Try just one stick of memory instead of two. Try slower FSB.
Prime95 is online now   Reply With Quote
Old 2005-01-16, 04:24   #3
cartmanea
 
Oct 2004
Corvallis, OR

2·5 Posts
Default

I'm getting the same results and my memory passes countless loops of Memtest86. I've had the problem for about 9 months now and its really driving me insane. I've tried re-downloading Prime95 too, thinking my copy was maybe corrupt, still a no-go. I've also tried 3 different motherboards, both amd and p4, and 4 cpu's, all with the same results. I've also tried lowering my fsb, which makes it last longer but it eventually fails.

Could it be my memory even though Memtest86 doesn't find any problems? Please help, I'm about bald from pulling all my hair out!
cartmanea is offline   Reply With Quote
Old 2005-01-16, 08:09   #4
ben
 
Jan 2005

11 Posts
Default

On SCO OpenServer 5.0.6, Pentium 4 Xeon 2.4GHz CPU, 4GBytes Memory.
I compile Mlucas with gcc-2.95.2:
$ gcc -o Mlucas -static -O3 *.c -lm
Then run Mlucas -s m also was having failings in large FFT test, but the Small FFT's doesn't.
If not optimize:
$ gcc -o Mlucas -static *.c -lm
Then all test pass, but speed is slow.
I compile Mlucas with gcc-3.4.3:
$ gcc -o Mlucas -static -O3 *.c -lm
Then all test pass, speed is fast.
see http://ben.skyiv.com/test/Mlucas.htm

Thus, this is relate to compiler and optimize.

Last fiddled with by ben on 2005-01-16 at 08:15
ben is offline   Reply With Quote
Old 2005-01-16, 16:16   #5
madbullet
 
Jan 2005

810 Posts
Default

hey cartmanea what Memtest are you using? im using memTest86+ version, which should an error (so MR prime was right aboutthe memory, give the man a drink ) i did use Memtest (non +) at first and didnt show any results.

if its like mine then it'll be the RAM (although i haven't been able to change it yet, all my test do indicate its a memory problem) so try the + version of memtest and see what happens.

thansk for all those who were kind enough to respond to my posts
madbullet is offline   Reply With Quote
Old 2005-01-16, 21:58   #6
cartmanea
 
Oct 2004
Corvallis, OR

2×5 Posts
Default

I used the latest version of Memtest86+
cartmanea is offline   Reply With Quote
Old 2005-01-17, 01:46   #7
madbullet
 
Jan 2005

23 Posts
Default

have you tried (if u can) a different stick of RAM? you say if tried different motherboards and CPU's so thats rules them out. change hardware one by one to see what the problem is, of course this is made alto easier if you have friends with spare parts.

good luck
madbullet is offline   Reply With Quote
Old 2005-01-17, 02:52   #8
cartmanea
 
Oct 2004
Corvallis, OR

10102 Posts
Default

I found the problem......

One of my sticks of BH-5 had a surface mount capacitor broken off. Its small and VERY hard to notice. Memtest86+ still fails to show any errors, even when I test that stick alone, but Prime95 points the problem out right away.

I guess that tells us all something about Memtest86

Last fiddled with by cartmanea on 2005-01-17 at 02:53
cartmanea is offline   Reply With Quote
Old 2005-01-17, 05:11   #9
geoff
 
geoff's Avatar
 
Mar 2003
New Zealand

115710 Posts
Default

Interesting. I guess memtest86 doesn't put enough load on the rest of the system to make this capicitor necessary.
geoff is offline   Reply With Quote
Old 2005-01-17, 06:22   #10
cartmanea
 
Oct 2004
Corvallis, OR

2·5 Posts
Default

I guess so, or maybe memtest86 doesn't store data in the memory for long periods of time and this capacitor helps keep the charge up so it maintains memory.
cartmanea is offline   Reply With Quote
Old 2005-01-17, 14:16   #11
madbullet
 
Jan 2005

23 Posts
Default

OO, well thats very interesting, ive been trying to find another Memory only stress test, but haven't found much, does anybody know if a good memory only stress test?

glad to see you found you problem .
madbullet is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Manual assignment results ? Have I found a new prime ? -- False alarm. JonRussell PrimeNet 21 2018-02-28 02:08
Strange benchmark results AlTonno15 Information & Answers 3 2013-01-29 02:23
Athlon II 620 x4 henryzz Hardware 5 2009-10-28 03:25
strange output in results.txt ("M5 is prime!")? joblack Information & Answers 9 2008-12-30 22:45
Empty results.txt/prime.log files on dual-core Graff Software 9 2007-05-08 15:31

All times are UTC. The time now is 21:30.

Wed Apr 14 21:30:33 UTC 2021 up 6 days, 16:11, 0 users, load averages: 2.04, 2.18, 2.24

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.