Prime 95 25.7 Crashing at 3072K FFT or greater

2008-11-05, 22:14

Hello,

I've used Prime95 for many years as a means of testing the stability of various builds and have in, the last few days encountered a very strange problem.

I have an 8 Core machine (2x QX9775 CPUs on an intel D5400XS Skulltrail motherboard) running Prime95 25.7 and have found that no matter what I try, when it attempts to run the 3072K or 4096 FFT test on all 8 cores, the program crashes to desktop before completing even one batch of iterations on any core,

I found this out the hard way by first running the program for just over 9 hours whereupon it crashed as soon as it reached this particular FFT size, hence I then manually set it to run from 2560K upwards and once again, it crashed very shortly after switching to 3072K.

Note that no rounding error is displayed on eny core, the application simply crashes to desktop with the ususal microsoft error report box.

I've tested using just 6 cores and this seems to be ok. 7 Cores workes sometimes but not others, 8 cores always seems to crash the program.

I've tried everyting I can possible think of in terms of voltage tweaking and swapping out of hardware (swapped out pretty much everyting apart from the CPUs). Temperatures all seem fine and system is otherwise rock solid.

Is this some sort of memory allocation bug or is it indeed a hardware fault.

If there is anyone who has a similar system around here that can run the same test and report back their findings, I'd be most grateful!!

jrk · 2008-11-06, 02:26

Either a bad memory module, or wrong timings.

Try adjusting timings and run memtest.

2008-11-06, 23:56

I must say I find that difficult to believe.

I've tried everything you can think of hardware wise, including running memtest (which passes multiple times without errors) and testing with each pair of memory modules seperately, with the other two removed from the system.

I've also, upped the voltage on the memory, loosened the timings, upped the Vcore on both CPUs, changed power supplies, tried with vista and xp (32bit and 64bit) and tested with the 64bit and 32bit versions of Prime.

I've also made another discovery, if I run a custom test and set Prime to use only 1550mb of memory or less, the problem never seems to occur, but with anything over 1600mb (the default setting for 25.6) it crashes to desktop.

So doesn't this mean it's more likely a software/memory allocation issue?

How can the system be perfectly stable for nine hours with every single test below 3072FFT, yet it can't complete even a single round at 3072K or above if the program is set to test more than 1600mb?????

Might it be something to do with running 8 cores and Prime thinking it doesn't have enough memory to work with?

rudi_m · 2008-11-08, 12:13

Did you used multithreaded mprime 25.x whith 8 threads?
So maybe you could try to run 8 mprime instances each using only 1 thread - just to see wheather this works or crashs too.

S485122 · 2008-11-08, 12:43

I understand that you are doing stress tests only, right ?

When you run the test on 6 cores with the memory to test set at 1800 MB, what is the total memory usage of Prime95 ?

How much physical memory do you have on the system ?

What OS are you using ?

Are you using the 64 bits version of the program or the 32 bits one ? (At the moment the 25,7 64 bits version will not work with more than 1702 MB, some users are eagerly waiting for a bug fix.)

Jacob

2008-12-17, 20:21

i am having the same problem as mentioned on this thread

i am also using the Prime 25.6 on windows xp 64 bit to do a blend test on a 8 core CPU with 8 GB memory installed ( 4GB per QUAD core CPU) . The test runs fine for 9-10 hrs but as soon as it reaches 3072K it either stops or gives a microsoft error

i have already tested the memory under memtest and have no issues with it

is it something of a software bug or should i decrease the memory allocation (set by default to 1750 MB) .let me know what i can do here

how can i decrease the CPU core from 8 to 6 as mentioned by someone before

thanks for all the help

Prime95 · 2008-12-18, 02:05

Does it crash immediately if you choose custom torture test with FFT sizes 3072K to 4096K? Does it crash if you run in-place FFTs? Does it crash if you run version 25.8?

It sounds like a software bug. Your help in isolating the trouble is greatly appreciated.

2008-11-05, 22:14	#1
Unregistered 14022₈ Posts	Prime 95 25.7 Crashing at 3072K FFT or greater Hello, I've used Prime95 for many years as a means of testing the stability of various builds and have in, the last few days encountered a very strange problem. I have an 8 Core machine (2x QX9775 CPUs on an intel D5400XS Skulltrail motherboard) running Prime95 25.7 and have found that no matter what I try, when it attempts to run the 3072K or 4096 FFT test on all 8 cores, the program crashes to desktop before completing even one batch of iterations on any core, I found this out the hard way by first running the program for just over 9 hours whereupon it crashed as soon as it reached this particular FFT size, hence I then manually set it to run from 2560K upwards and once again, it crashed very shortly after switching to 3072K. Note that no rounding error is displayed on eny core, the application simply crashes to desktop with the ususal microsoft error report box. I've tested using just 6 cores and this seems to be ok. 7 Cores workes sometimes but not others, 8 cores always seems to crash the program. I've tried everyting I can possible think of in terms of voltage tweaking and swapping out of hardware (swapped out pretty much everyting apart from the CPUs). Temperatures all seem fine and system is otherwise rock solid. Is this some sort of memory allocation bug or is it indeed a hardware fault. If there is anyone who has a similar system around here that can run the same test and report back their findings, I'd be most grateful!!

2008-12-17, 20:21	#6
Unregistered 3×5×389 Posts	Prime 95 25.6 crashing at 3072K FFT or greater i am having the same problem as mentioned on this thread i am also using the Prime 25.6 on windows xp 64 bit to do a blend test on a 8 core CPU with 8 GB memory installed ( 4GB per QUAD core CPU) . The test runs fine for 9-10 hrs but as soon as it reaches 3072K it either stops or gives a microsoft error i have already tested the memory under memtest and have no issues with it is it something of a software bug or should i decrease the memory allocation (set by default to 1750 MB) .let me know what i can do here how can i decrease the CPU core from 8 to 6 as mentioned by someone before thanks for all the help

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
All mersenne primes greater than 3 will end with either 1 or 7?	soumya	Miscellaneous Math	2	2013-03-28 21:33
LLR Crashing	mdettweiler	Riesel Prime Search	14	2007-09-30 18:29
Prime 95 crashing on laptop running as service	remaker	Software	2	2007-03-19 02:26
V24.12 Release 1 Crashing on P4 M	njcroquet1	Software	8	2005-06-24 14:40
Prime crashing	russ	Software	3	2002-08-29 00:08

2008-11-06, 02:26	#2
jrk May 2008 10001000111₂ Posts	Either a bad memory module, or wrong timings. Try adjusting timings and run memtest.

2008-11-06, 23:56	#3
Unregistered 514₈ Posts	I must say I find that difficult to believe. I've tried everything you can think of hardware wise, including running memtest (which passes multiple times without errors) and testing with each pair of memory modules seperately, with the other two removed from the system. I've also, upped the voltage on the memory, loosened the timings, upped the Vcore on both CPUs, changed power supplies, tried with vista and xp (32bit and 64bit) and tested with the 64bit and 32bit versions of Prime. I've also made another discovery, if I run a custom test and set Prime to use only 1550mb of memory or less, the problem never seems to occur, but with anything over 1600mb (the default setting for 25.6) it crashes to desktop. So doesn't this mean it's more likely a software/memory allocation issue? How can the system be perfectly stable for nine hours with every single test below 3072FFT, yet it can't complete even a single round at 3072K or above if the program is set to test more than 1600mb????? Might it be something to do with running 8 cores and Prime thinking it doesn't have enough memory to work with?

2008-11-08, 12:13	#4
rudi_m Jul 2005 182₁₀ Posts	Did you used multithreaded mprime 25.x whith 8 threads? So maybe you could try to run 8 mprime instances each using only 1 thread - just to see wheather this works or crashs too.

2008-11-08, 12:43	#5
S485122 "Jacob" Sep 2006 Brussels, Belgium 2×977 Posts	I understand that you are doing stress tests only, right ? When you run the test on 6 cores with the memory to test set at 1800 MB, what is the total memory usage of Prime95 ? How much physical memory do you have on the system ? What OS are you using ? Are you using the 64 bits version of the program or the 32 bits one ? (At the moment the 25,7 64 bits version will not work with more than 1702 MB, some users are eagerly waiting for a bug fix.) Jacob

2008-12-18, 02:05	#7
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 8279₁₀ Posts	Does it crash immediately if you choose custom torture test with FFT sizes 3072K to 4096K? Does it crash if you run in-place FFTs? Does it crash if you run version 25.8? It sounds like a software bug. Your help in isolating the trouble is greatly appreciated.