Round off errors

Matt_G · 2004-04-04, 16:04

Hi Guys,
I'm having a problem with a P4 3.0 Northwood on a Intel 875PBZ motherboard.
(No overclock)
Every other LL test it is throwing round off errors.
Example from yesterday morning:
[Sat Apr 03 09:54:47 2004]
Iteration: 9577984/2167xxxx, ERROR: ROUND OFF (0.5) > 0.40
Continuing from last save file.
[Sat Apr 03 12:18:49 2004]
Iteration: 9721600/2167xxxx, ERROR: ROUND OFF (0.5) > 0.40
Continuing from last save file.

I save checkpoint files every 1M iterations. The residues every 1M are in results.txt
So when I saw what had happened this morning, I restarted the crunch from the save file at 9M. I will check the residues at 10M to see if they match. (It had got up to 11.2M iterations before I caught it.)

Will heat cause this?
I'm pretty sure the answer is yes. Right now with Prime running the processor is at 56C, according to Motherboard Monitor. I will run it with the side off for a while. It's already maxed out on fans barring cutting holes in the case.

Should I turn on round off checking? Would that help? How much of a hit would it cause? Not that that matters much, if that is what it takes to get good results, so be it.

Matt

Matt_G · 2004-04-04, 16:32

Update:
I took the side off the comp and blew the dust out of my Zalman 7000AlCu heatsink (I just did that a month ago...sheesh) and the temp on the chip dropped to 48C. Will see how it does.....

Matt_G · 2004-04-10, 14:27

No errors since my post above. Seems it was a heat problem.

I cut the error related text out of my results.txt
I'm pretty sure this will cause the error code submitted with the residue to be 00000000, which technically is correct since I restarted the crunch from a time before the errors.
Should I put it back in, or leave it??

dsouza123 · 2004-04-11, 07:54

The results.txt file isn't sent to primenet so it doesn't matter what it shows.
It is a binary prime.spl file that gets sent.

The prime.spl is sent for trial factoring results, can someone confirm it is also
for LL and DC and P-1 factoring.

For a manually reserved exponent (from the web page) the results are the final couple of lines
of results from the results.txt file for the exponent, not all the lines about the exponent.

Danath · 2004-04-12, 14:46

When using PrimeNet, prime.spl is created and sent for all Prime95 tasks: trial-factoring, double-checking, P-1 factoring and LL-testing, just after every single task is completed.

I'm confident that every time an error occurs, it's recorded into the exponent's savefile, so if you took a savefile that was saved before the first error happened, and there won't be other errors, the final result should be clean (with no errors).

2004-04-04, 16:04	#1
Matt_G Jan 2004 Dacono, CO. 2³ Posts	Round off errors Hi Guys, I'm having a problem with a P4 3.0 Northwood on a Intel 875PBZ motherboard. (No overclock) Every other LL test it is throwing round off errors. Example from yesterday morning: [Sat Apr 03 09:54:47 2004] Iteration: 9577984/2167xxxx, ERROR: ROUND OFF (0.5) > 0.40 Continuing from last save file. [Sat Apr 03 12:18:49 2004] Iteration: 9721600/2167xxxx, ERROR: ROUND OFF (0.5) > 0.40 Continuing from last save file. I save checkpoint files every 1M iterations. The residues every 1M are in results.txt So when I saw what had happened this morning, I restarted the crunch from the save file at 9M. I will check the residues at 10M to see if they match. (It had got up to 11.2M iterations before I caught it.) Will heat cause this? I'm pretty sure the answer is yes. Right now with Prime running the processor is at 56C, according to Motherboard Monitor. I will run it with the side off for a while. It's already maxed out on fans barring cutting holes in the case. Should I turn on round off checking? Would that help? How much of a hit would it cause? Not that that matters much, if that is what it takes to get good results, so be it. Matt Last fiddled with by Matt_G on 2004-04-04 at 16:05

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Round off error	Androx72	Software	2	2013-02-28 00:00
Large Round Off Errors	evoflash	Software	8	2013-02-10 18:39
20 Questions - round 5	retina	Puzzles	110	2010-11-10 15:26
Benchmarks please - 25.9 (round 2)	Prime95	Software	80	2009-03-15 12:13
Hardware, FFT limits and round off errors	ewergela	Hardware	9	2005-09-01 14:51

2004-04-04, 16:32	#2
Matt_G Jan 2004 Dacono, CO. 2³ Posts	Update: I took the side off the comp and blew the dust out of my Zalman 7000AlCu heatsink (I just did that a month ago...sheesh) and the temp on the chip dropped to 48C. Will see how it does.....

2004-04-10, 14:27	#3
Matt_G Jan 2004 Dacono, CO. 2³ Posts	No errors since my post above. Seems it was a heat problem. I cut the error related text out of my results.txt I'm pretty sure this will cause the error code submitted with the residue to be 00000000, which technically is correct since I restarted the crunch from a time before the errors. Should I put it back in, or leave it??

2004-04-11, 07:54	#4
dsouza123 Sep 2002 2·331 Posts	The results.txt file isn't sent to primenet so it doesn't matter what it shows. It is a binary prime.spl file that gets sent. The prime.spl is sent for trial factoring results, can someone confirm it is also for LL and DC and P-1 factoring. For a manually reserved exponent (from the web page) the results are the final couple of lines of results from the results.txt file for the exponent, not all the lines about the exponent.

2004-04-12, 14:46	#5
Danath May 2003 Republic of Moldova 2³·5 Posts	When using PrimeNet, prime.spl is created and sent for all Prime95 tasks: trial-factoring, double-checking, P-1 factoring and LL-testing, just after every single task is completed. I'm confident that every time an error occurs, it's recorded into the exponent's savefile, so if you took a savefile that was saved before the first error happened, and there won't be other errors, the final result should be clean (with no errors).