Small FFTs vs. Blend Crash Problem

MoeStooge · 2003-12-06, 05:36

I was trying out the torture test with a new PC I'm building and for some reason Prime95 crashes within 2-4 hours if I perform the blend test, but will run indefinitely with the small FFT test. It never gives me any type of calculation error, it just crashes. Originally, I thought that I had a memory problem but after replacing it I am still getting the same results. Memtest86 runs perfectly and all my normal applications run flawlessly. Other than memory, what would be involved in the blend test that could give me a hint where the problem is?

btw, this is a P4 2.6C system running with default settings (no overclocking).

Schrottie · 2003-12-06, 14:11

Hi. Same behaviour on my new PC. I've got a P42.8 with hyperthreading, an Asus P4C800-E and 2x512 MB Kingston valueram. Also no Overclocking.
Windows (Xp SP1) reports that Prime95 has encountered a problem and wants to report it to Microsoft.
This happens always after 3 hours or so.
I made several new Windows installations and also tested slower RAM-timings in BIOS.
The only card installed is my Geforce4 (Asus 8460Ultra).
I've disabled raid,sound and all legacy device support.
The only devices attached are one IDE harddrive and one DVD-ROM.
I've tested directly after a clean Windows install (Inf update installed of course) and Prime always crashes.
I've checked temperature also (Cpu max 48deg.C ; Board 37deg.C) I think that's ok.

The last time i tried running the "blend-test" , I realised that Prime95 seems to crash always at the same point, when testing:
270000 iterations of M420217 using 20K FFT length
I will inspect that further.
Perhaps someone can tell me what I have to configure to repeat
a specific iteration without doing the whole torture test.

Perhaps we have found a new Pentium-bug....

MrHappy · 2003-12-06, 14:24

"Board 37deg.C"
that's quite a lot I think.

Schrottie · 2003-12-06, 14:40

Quote:

Originally posted by MrHappy
"Board 37deg.C"
that's quite a lot I think.

Sorry. Information was incorrect. I checked now while running the torture test. Cpu 46; MB 33. I checked this with Asus Pc-Probe.

But I already did a torture test with open PC-case (MB temp was 27 or 28) and got the same crash.

There's no message within Prime95 that there's a false claculation.
It just crashes.

dsouza123 · 2003-12-06, 15:19

It maybe some other program/driver has a bug that effects the calculation.

Try running XP in safe mode it uses different drivers and loads less programs and drivers.

A link to running XP in safe mode
http://service1.symantec.com/SUPPORT...01060608000039

MoeStooge · 2003-12-06, 18:09

Quote:

Originally posted by Schrottie
The last time i tried running the "blend-test" , I realised that Prime95 seems to crash always at the same point, when testing:
270000 iterations of M420217 using 20K FFT length
I will inspect that further.
Perhaps someone can tell me what I have to configure to repeat
a specific iteration without doing the whole torture test.

Perhaps we have found a new Pentium-bug....

Wow. That's exactly where it happened for me.

It was on M420217 with the 20K FFT.

I tried custom mode and put in 20K for the max and min FFT and it happens every time. I altered the memory setting from its default of 926MB to 800MB and it passed through without a problem though. Unfortunately, it has moved to another exponent so I can't try it again. I wonder how you set the exponent value that it starts with in the custom test. Right now I am running it again and hoping that it cycles back through to M420217.

The fact that it is happening in the exact same spot makes me feel better about my situation at least.

MrHappy · 2003-12-06, 18:37

I guess Prime95 doesn't work correctly with very small FFTs.

Schrottie · 2003-12-06, 19:10

Quote:

Originally posted by MoeStooge
Wow. That's exactly where it happened for me.

It was on M420217 with the 20K FFT.
...

The fact that it is happening in the exact same spot makes me feel better about my situation at least.

Hi again!

I tested in safe mode according to the post from dsouza123.
And it crashes again at M420217 with 20K FFT !!!
I've also set min and max to 20K in custom mode and Prime95 crashes directly.
With memory set to 800MB it seems to run well.
Also 925MB and 927MB is working.
Can someone explain that??

MoeStooge, which mainboard you have?
And what RAM configuration? (I've got 2x512MB)
Perhaps this is important, too.

Either there is a general problem with the new P4 or George Woltman has to fix a bug in his sourcecode.

MoeStooge · 2003-12-06, 19:18

I'm using an ASUS P4C800E with 2x512MB. The memory timings don't seem to affect the problem either.

On another messageboard:
http://forums.anandtech.com/messagev...E=30&zb=322999

Someone else is having the same problem.

Did you figure out a way of hand choosing the exponent?

Schrottie · 2003-12-06, 19:35

Quote:

Originally posted by MoeStooge
I'm using an ASUS P4C800E with 2x512MB. The memory timings don't seem to affect the problem either.

On another messageboard:
http://forums.anandtech.com/messagev...E=30&zb=322999

Someone else is having the same problem.

Did you figure out a way of hand choosing the exponent?

No, I didn't find a way to.
This problem seems to be MB-specific, because I've got a P4C800E, too.

Schrottie · 2003-12-06, 19:55

I tested again the 926MB with 20K but now I've removed one DIMM.
So the BIOS reported single channel mode.
First it looked like Prime is running now, but after a minute the harddisk started heavy action due to memoryswapping I think.

Because of having physically 512MB Windows has to use the pagefile now to reach 926MB.
And after 10 seconds harddisk action Prime crashed again.
So I think RAM size and configuration is not the reason for this.

2003-12-06, 05:36	#1
MoeStooge Dec 2003 3² Posts	Small FFTs vs. Blend Crash Problem I was trying out the torture test with a new PC I'm building and for some reason Prime95 crashes within 2-4 hours if I perform the blend test, but will run indefinitely with the small FFT test. It never gives me any type of calculation error, it just crashes. Originally, I thought that I had a memory problem but after replacing it I am still getting the same results. Memtest86 runs perfectly and all my normal applications run flawlessly. Other than memory, what would be involved in the blend test that could give me a hint where the problem is? btw, this is a P4 2.6C system running with default settings (no overclocking).

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Small FFTs immediately crashes my computer, help please!	scrawlings	Information & Answers	39	2014-08-02 21:48
Weird error message in small fft and blend	samhot84	Information & Answers	4	2014-04-20 19:40
P95 PrimeNet causes BSOD; small FFTs, large FFTs, and blend test don't	KarateF22	PrimeNet	16	2013-10-28 00:34
blend / small FFT torture test on linux ?	lukasz	Linux	1	2007-01-09 16:16
Windows XP BSOD Prime95 Blend Test - Paging Problem???	john6205	Software	18	2006-11-20 17:21

2003-12-06, 14:11	#2
Schrottie Dec 2003 7 Posts	Hi. Same behaviour on my new PC. I've got a P42.8 with hyperthreading, an Asus P4C800-E and 2x512 MB Kingston valueram. Also no Overclocking. Windows (Xp SP1) reports that Prime95 has encountered a problem and wants to report it to Microsoft. This happens always after 3 hours or so. I made several new Windows installations and also tested slower RAM-timings in BIOS. The only card installed is my Geforce4 (Asus 8460Ultra). I've disabled raid,sound and all legacy device support. The only devices attached are one IDE harddrive and one DVD-ROM. I've tested directly after a clean Windows install (Inf update installed of course) and Prime always crashes. I've checked temperature also (Cpu max 48deg.C ; Board 37deg.C) I think that's ok. The last time i tried running the "blend-test" , I realised that Prime95 seems to crash always at the same point, when testing: 270000 iterations of M420217 using 20K FFT length I will inspect that further. Perhaps someone can tell me what I have to configure to repeat a specific iteration without doing the whole torture test. Perhaps we have found a new Pentium-bug....

2003-12-06, 14:24	#3
MrHappy Dec 2003 Paisley Park & Neverland B9₁₆ Posts	"Board 37deg.C" that's quite a lot I think.

2003-12-06, 15:19	#5
dsouza123 Sep 2002 1010010110₂ Posts	It maybe some other program/driver has a bug that effects the calculation. Try running XP in safe mode it uses different drivers and loads less programs and drivers. A link to running XP in safe mode http://service1.symantec.com/SUPPORT...01060608000039

2003-12-06, 18:37	#7
MrHappy Dec 2003 Paisley Park & Neverland 5×37 Posts	I guess Prime95 doesn't work correctly with very small FFTs.

2003-12-06, 19:18	#9
MoeStooge Dec 2003 3² Posts	I'm using an ASUS P4C800E with 2x512MB. The memory timings don't seem to affect the problem either. On another messageboard: http://forums.anandtech.com/messagev...E=30&zb=322999 Someone else is having the same problem. Did you figure out a way of hand choosing the exponent?

2003-12-06, 19:55	#11
Schrottie Dec 2003 7 Posts	I tested again the 926MB with 20K but now I've removed one DIMM. So the BIOS reported single channel mode. First it looked like Prime is running now, but after a minute the harddisk started heavy action due to memoryswapping I think. Because of having physically 512MB Windows has to use the pagefile now to reach 926MB. And after 10 seconds harddisk action Prime crashed again. So I think RAM size and configuration is not the reason for this.