Large Round Off Errors

evoflash · 2012-12-05, 09:27

Hello everyone :-)

Roughly since I've been using Windows 7, I've started seeing roundoff errors.

Big ones.

Like this:

Code:

[Wed Aug 01 13:01:55 2012]
Iteration: 28144640/52986251, ERROR: ROUND OFF (8.163158034e+140) > 0.40
Continuing from last save file.
[Wed Aug 15 17:37:41 2012]
Iteration: 49287424/52986251, ERROR: ROUND OFF (6.04438439e+232) > 0.40
Continuing from last save file.
[Thu Aug 16 19:21:22 2012]
Iteration: 52783488/52986251, ERROR: ROUND OFF (4.883952185e+257) > 0.40
Continuing from last save file.
Iteration: 52783480/52986251, ERROR: ROUND OFF (4.883952185e+257) > 0.40
Continuing from last save file.
[Thu Aug 16 20:05:08 2012]
UID: evoflash/Work_Laptop_3, M52986251 is not prime. Res64: EC2C21EFDB23D651. We4: 0A94714E,33693360,00000400, AID: F2DD985B79377CAC0300D6F2544475C9
[Thu Oct 11 09:51:28 2012]
UID: evoflash/Work_Laptop_3, M55584869 is not prime. Res64: 9B4D2985382B0335. We4: 7C8D2773,8298347,00000000, AID: 452C3E74CC8E7AE8C592CF4A978E8666
[Wed Oct 17 10:00:33 2012]
Iteration: 14877440/29585693, ERROR: ROUND OFF (8.163158034e+140) > 0.40
Continuing from last save file.
Iteration: 14877432/29585693, ERROR: ROUND OFF (8.163158034e+140) > 0.40
Continuing from last save file.
Iteration: 14877424/29585693, ERROR: ROUND OFF (8.163158034e+140) > 0.40
Continuing from last save file.
[Tue Oct 23 11:22:50 2012]
UID: evoflash/Work_Laptop_3, M29585693 is not prime. Res64: EEA561420417177C. We4: 6DE35886,7708248,00000300, AID: BBF74CDE088C5CFA5ED77806E1EF8E98
[Wed Nov 28 08:46:21 2012]
Iteration: 23006976/57440087, ERROR: ROUND OFF (1.142615744e+243) > 0.40
Continuing from last save file.
Iteration: 23006968/57440087, ERROR: ROUND OFF (1.142615744e+243) > 0.40
Continuing from last save file.
Iteration: 23006960/57440087, ERROR: ROUND OFF (1.142615744e+243) > 0.40
Continuing from last save file.
Iteration: 23006952/57440087, ERROR: ROUND OFF (1.142615744e+243) > 0.40
Continuing from last save file.
Iteration: 23006944/57440087, ERROR: ROUND OFF (1.142615744e+243) > 0.40
Continuing from last save file.
Iteration: 23006936/57440087, ERROR: ROUND OFF (1.142615744e+243) > 0.40
Continuing from last save file.

I get this on my home PC too. Thing is, I'm constantly telling Win7 to force shutdown because it takes forever to close programs. Anyone think it's connected? Or do I have two sickly CPU's? (Both <1yr old).

Appreciate any comment.

garo · 2012-12-05, 11:28

Check the CPU temperature. if it is an Intel use CoreTemp. Also check to see if dust is clogging your air intake and/or heatsink.

If you are overclocking the machine it is possible the overclock is too high.

fivemack · 2012-12-05, 12:05

This looks like a software bug to me; those differences are ASCII strings!

Code:

8.16316e+140    5d3123207264d5f6        ��dr #1]
6.04438e+232    704377654e063d44        D=NewCp
4.88395e+257    75704377654a6689        �fJewCpu
1.14262e+243    72656b726f47d952        R�Gorker

(OK, they're not given to full precision so we only get the last four characters of the string, but those definitely aren't single bits flipped by dodgy FP hardware!)

Dubslow · 2012-12-05, 20:00

Yes, fivemack's right, no roundoff should be greater than 0.500; if it is, then there's a software error.

This is a known issue for certain (older) versions of Prime95 v27. What version are you using? If it is v27, be sure that it's 27.7 build 2 or later, and if not, upgrade and try again.

LaurV · 2012-12-06, 02:54

Quote:

Originally Posted by fivemack

This looks like a software bug to me; those differences are ASCII strings!

Code:

8.16316e+140    5d3123207264d5f6        ��dr #1]
6.04438e+232    704377654e063d44        D=NewCp
4.88395e+257    75704377654a6689        �fJewCpu
1.14262e+243    72656b726f47d952        R�Gorker

(OK, they're not given to full precision so we only get the last four characters of the string, but those definitely aren't single bits flipped by dodgy FP hardware!)

Very good - genuine ingenious, as one colleague of mine would say - catch.

My money are on temperature thingies, causing some bad jumps. For any windows version, RealTemp would be a start, or some better tool with a history or graphic representation.

evoflash · 2013-01-30, 13:19

Removal and reinstall has seen no further errors crop up. My thanks to you :)

99.20% LL on M57440087 :)

Prime95 · 2013-01-30, 13:47

This bug, which occurs only when running a multi-threaded test, has been fixed in 27.9. The good news is the bug does not affect the validity of your result.

mwn3d · 2013-02-10, 15:26

When is 27.9 being released?

kladner · 2013-02-10, 18:39

Quote:

Originally Posted by mwn3d

When is 27.9 being released?

It is already available at http://www.mersenne.org/freesoft/ and on various mirrors. The name is p95v279.win64.zip ......EDIT:

if you are running 64 bit Windows. All the other OS versions are there, too.

2012-12-05, 12:05	#3
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 2·7·461 Posts	This looks like a software bug to me; those differences are ASCII strings! Code: 8.16316e+140 5d3123207264d5f6 ��dr #1] 6.04438e+232 704377654e063d44 D=NewCp 4.88395e+257 75704377654a6689 �fJewCpu 1.14262e+243 72656b726f47d952 R�Gorker (OK, they're not given to full precision so we only get the last four characters of the string, but those definitely aren't single bits flipped by dodgy FP hardware!) Last fiddled with by fivemack on 2012-12-05 at 12:07

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
20 Questions Round 8	Uncwilly	Puzzles	41	2010-12-20 18:53
40 questions round 3	TimSorbet	Puzzles	56	2010-12-08 13:08
20 Questions - round 7	Flatlander	Puzzles	52	2010-12-01 12:52
Hardware, FFT limits and round off errors	ewergela	Hardware	9	2005-09-01 14:51
Round off errors	Matt_G	Hardware	4	2004-04-12 14:46

2012-12-05, 11:28	#2
garo Aug 2002 Termonfeckin, IE AD0₁₆ Posts	Check the CPU temperature. if it is an Intel use CoreTemp. Also check to see if dust is clogging your air intake and/or heatsink. If you are overclocking the machine it is possible the overclock is too high.

2012-12-05, 20:00	#4
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	Yes, fivemack's right, no roundoff should be greater than 0.500; if it is, then there's a software error. This is a known issue for certain (older) versions of Prime95 v27. What version are you using? If it is v27, be sure that it's 27.7 build 2 or later, and if not, upgrade and try again.

2013-01-30, 13:19	#6
evoflash Dec 2012 2²×7 Posts	Removal and reinstall has seen no further errors crop up. My thanks to you :) 99.20% LL on M57440087 :)

2013-01-30, 13:47	#7
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2²×2,089 Posts	This bug, which occurs only when running a multi-threaded test, has been fixed in 27.9. The good news is the bug does not affect the validity of your result.

2013-02-10, 15:26	#8
mwn3d Feb 2013 1₈ Posts	When is 27.9 being released?