Prime95 30.8 (beta, big P-1 changes, see post #551) - Page 53

pepi37 · 2022-05-21, 22:52

Prime95 v 30.8 build 15
Intel I5 9600K
Win 10

Input line
Pminus1=286037,2,52635,1,900000,5000000
Output result
286037*2^52635+1 completed P-1, B1=900000, B2=24612017430, Wi4: BA8D52BA

I know that was huge changes and speed up in P-1 but never before B2 was higher then I request in input file.
If it is some new behavior then I apologize.

Prime95 · 2022-05-22, 00:30

Quote:

Originally Posted by pepi37

[May 21 13:38:17] Error setting affinity to cpuset 0x00000020: No error
And with same settings benchmark on 400K dont even start

I cannot reproduce this problem. Has anyone else seen benchmarking issues with build 15? Does the hwloc output in results.bench.txt look reasonable?

Quote:

Originally Posted by pepi37

I know that was huge changes and speed up in P-1 but never before B2 was higher then I request in input file. If it is some new behavior then I apologize.

This is indeed new behavior. I haven't decided if there should be an option to more closely obey the bounds specified in worktodo.txt. Right now prime95 uses all the memory available and chooses the first convenient B2 larger than the one specified in worktodo.txt.

DrobinsonPE · 2022-05-22, 04:08

I was out last night and today. Came home, and it looked like one of my computers had skipped turning in an assignment. Except, the results.json.txt file on the computer showed a result. The computer has turned in six other assignments after this particular one and all of them have been accepted. Thinking that possibly there was just a communication glitch, I manually turned in (copy and pasted) the result from results.json.txt.

the assignment was Pminus1=N/A,1,2,3885489,-1,23400000,0,71

below is what PrimeNet returned. Does this indicate that there was an error during the run? I still have the data file for the exponent and the result from results.json.txt. Unfortunately, the terminal window does not scroll back far enough to see this exponent.

Code:

Found 2 lines to process at 2022-05-22T03:46:16
Splitting composite factor 150961815138972606791943258786857465689505921666311960198815351986895087511287875524151819307631 into:
* 7
* 73
* 2593
* 262657
* 71119
* 97685839
* 2789781103
* 159161143
* 404090857
* 29978287182817
* 6465453697
* 38685655369
* 46408280617
processing: P-1 factor 7 for M3885489 (B1=23,400,000, B2=23,400,000) (2.807 bits)
Error code: 40, error text: Factor 7 reported for untracked M3885489
processing: P-1 factor 73 for M3885489 (B1=23,400,000, B2=23,400,000) (6.190 bits)
Error code: 40, error text: Factor 73 reported for untracked M3885489
processing: P-1 factor 2593 for M3885489 (B1=23,400,000, B2=23,400,000) (11.340 bits)
Error code: 40, error text: Factor 2593 reported for untracked M3885489
processing: P-1 factor 262657 for M3885489 (B1=23,400,000, B2=23,400,000) (18.003 bits)
Error code: 40, error text: Factor 262657 reported for untracked M3885489
processing: P-1 factor 71119 for M3885489 (B1=23,400,000, B2=23,400,000) (16.118 bits)
Error code: 40, error text: Factor 71119 reported for untracked M3885489
processing: P-1 factor 97685839 for M3885489 (B1=23,400,000, B2=23,400,000) (26.542 bits)
Error code: 40, error text: Factor 97685839 reported for untracked M3885489
processing: P-1 factor 2789781103 for M3885489 (B1=23,400,000, B2=23,400,000) (31.378 bits)
Error code: 40, error text: Factor 2789781103 reported for untracked M3885489
processing: P-1 factor 159161143 for M3885489 (B1=23,400,000, B2=23,400,000) (27.246 bits)
Error code: 40, error text: Factor 159161143 reported for untracked M3885489
processing: P-1 factor 404090857 for M3885489 (B1=23,400,000, B2=23,400,000) (28.590 bits)
Error code: 40, error text: Factor 404090857 reported for untracked M3885489
processing: P-1 factor 29978287182817 for M3885489 (B1=23,400,000, B2=23,400,000) (44.769 bits)
Error code: 40, error text: Factor 29978287182817 reported for untracked M3885489
processing: P-1 factor 6465453697 for M3885489 (B1=23,400,000, B2=23,400,000) (32.590 bits)
Error code: 40, error text: Factor 6465453697 reported for untracked M3885489
processing: P-1 factor 38685655369 for M3885489 (B1=23,400,000, B2=23,400,000) (35.171 bits)
Error code: 40, error text: Factor 38685655369 reported for untracked M3885489
processing: P-1 factor 46408280617 for M3885489 (B1=23,400,000, B2=23,400,000) (35.434 bits)
Error code: 40, error text: Factor 46408280617 reported for untracked M3885489

axn · 2022-05-22, 04:41

3885489 = 3^4* 47969 is not a prime, so gimps doesn't track this number. Clearly something went wrong in the assignment itself.

The closest matching prime exponent with no factors is 3885487

DrobinsonPE · 2022-05-22, 05:03

I fat fingered the assignment. That explains so much. Thanks for pointing out the obvious answer. That was a GIGO exercise. Time to try again with the right number.

S485122 · 2022-05-22, 10:28

P-1 factoring will find a factor q=2*k*p-1, if k has just one factor between B1 and B2 and all remaining factors are below B1. Since 30.8 uses huge values for B2, one will miss those factors q that have more than one factor of K between the bounds. This was less of a problem when B2 was about 30 times B1.

Would it be possible and not cost too much time for the program to check a few times for factors during the stage 2 run instead of doing it at the end ?

axn · 2022-05-22, 15:58

Quote:

Originally Posted by S485122

P-1 factoring will find a factor q=2*k*p-1

2*k*p+1

Quote:

Originally Posted by S485122

Since 30.8 uses huge values for B2, one will miss those factors q that have more than one factor of K between the bounds.

Yes, but not specific to 30.8

Quote:

Originally Posted by S485122

This was less of a problem when B2 was about 30 times B1.

It is not so much a problem as it is a limitation of the P-1 algorithm.

Quote:

Originally Posted by S485122

Would it be possible and not cost too much time for the program to check a few times for factors during the stage 2 run instead of doing it at the end ?

For smaller exponents, this shouldn't cost much. But how would this help with the aforementioned problem? Or is it an independent suggestion that has nothing to do with the problem?

James Heinrich · 2022-05-22, 16:03

Quote:

Originally Posted by axn

But how would this help with the aforementioned problem?

Presumably on the assumption that stage-2 is done to incrementally-larger bounds as the stage progresses. Stage-1 bounds can be repeatedly extended from a previous result, but I'm not sure if stage-2 functions that way.

axn · 2022-05-22, 16:48

Quote:

Originally Posted by James Heinrich

Presumably on the assumption that stage-2 is done to incrementally-larger bounds as the stage progresses.

It is. But the essential nature of stage 2 remains, i.e. k must have one factor between B1 & B2. That doesn't change just because you checked for the factor (i.e. did a gcd) multiple times during stage 2. I still don't understand how that's meant to find factors where k has more than 1 factor between B1 & B2.

James Heinrich · 2022-05-22, 17:47

Quote:

Originally Posted by axn

I still don't understand how that's meant to find factors where k has more than 1 factor between B1 & B2.

I think the idea is that instead of running GCD once at the end of (B2=B1*100) it could be run multiple times as B2 is expanded, for example run it 10 times at successively higher B2 values (B2=B1*10*[1..10]).
That's how I interpret S485122's suggestion anyways.

That may allow you to find the stage-2 factor sooner (if less than the target B2 is required), but unless B1 is extended to above that factor then you still have the problem of more than 1 factor between B1 and B2.

James Heinrich · 2022-05-22, 18:35

I have an occasional issue of Prime95 loading a single core when it should be idle. I have PauseWhileRunning set, and right now due to another process running Prime95 has been (correctly) paused for 4 hours. But I see the Communication Thread shows as "active", prime95.exe is taking up 100% of a single core, but without any on-screen explanation (or in logs for that matter). Perhaps it's recalculating estimated completion dates? I'm not sure. Whatever it's doing, I don't think it should be doing while paused due to PauseWhileRunning (kind of defeats the purpose of the setting). Also, whatever it's doing in the communication thread should show up as an entry there (perhaps in prime.log as well).

2022-05-22, 10:28	#578
S485122 "Jacob" Sep 2006 Brussels, Belgium 1,823 Posts	Huge B2 might imply missing factors P-1 factoring will find a factor q=2kp-1, if k has just one factor between B1 and B2 and all remaining factors are below B1. Since 30.8 uses huge values for B2, one will miss those factors q that have more than one factor of K between the bounds. This was less of a problem when B2 was about 30 times B1. Would it be possible and not cost too much time for the program to check a few times for factors during the stage 2 run instead of doing it at the end ?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Prime95 beta version 28.4	Prime95	Software	20	2014-03-02 02:51
Prime95 beta version 28.3	Prime95	Software	68	2014-02-23 05:42
Prime95 version 27.1 early preview, not-even-close-to-beta release	Prime95	Software	126	2012-02-09 16:17
Beta version 24.12 available	Prime95	Software	33	2005-06-14 13:19
Beta version of PRP	Prime95	PSearch	15	2004-09-17 19:21

2022-05-21, 22:52	#573
pepi37 Dec 2011 After milion nines:) 2²×383 Posts	Prime95 v 30.8 build 15 Intel I5 9600K Win 10 Input line Pminus1=286037,2,52635,1,900000,5000000 Output result 2860372^52635+1 completed P-1, B1=900000, B2=24612017430*, Wi4: BA8D52BA I know that was huge changes and speed up in P-1 but never before B2 was higher then I request in input file. If it is some new behavior then I apologize.

2022-05-22, 04:41	#576
axn Jun 2003 2⁸×3×7 Posts	3885489 = 3^4* 47969 is not a prime, so gimps doesn't track this number. Clearly something went wrong in the assignment itself. The closest matching prime exponent with no factors is 3885487

2022-05-22, 05:03	#577
DrobinsonPE Aug 2020 141₁₀ Posts	I fat fingered the assignment. That explains so much. Thanks for pointing out the obvious answer. That was a GIGO exercise. Time to try again with the right number.

2022-05-22, 18:35	#583
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 3,733 Posts	I have an occasional issue of Prime95 loading a single core when it should be idle. I have PauseWhileRunning set, and right now due to another process running Prime95 has been (correctly) paused for 4 hours. But I see the Communication Thread shows as "active", prime95.exe is taking up 100% of a single core, but without any on-screen explanation (or in logs for that matter). Perhaps it's recalculating estimated completion dates? I'm not sure. Whatever it's doing, I don't think it should be doing while paused due to PauseWhileRunning (kind of defeats the purpose of the setting). Also, whatever it's doing in the communication thread should show up as an entry there (perhaps in prime.log as well).