mersenneforum.org > Data P-1 / P+1 / ECM strategy for PRP-CF
 Register FAQ Search Today's Posts Mark Forums Read

 2021-05-08, 15:29 #78 ATH Einyen     Dec 2003 Denmark 7·11·41 Posts I think RDS mentioned once regarding the old ECM paper he wrote, that for ECM ideally time spent in step1 and step2 should be the same, and in some ideal implementation B2=c*B12. But that was for ECM, not sure if that applied to P-1 and P+1. Last fiddled with by ATH on 2021-05-08 at 15:30
2021-05-08, 15:44   #79
alpertron

Aug 2002
Buenos Aires, Argentina

2×683 Posts

Quote:
 Originally Posted by ATH I think RDS mentioned once regarding the old ECM paper he wrote, that for ECM ideally time spent in step1 and step2 should be the same, and in some ideal implementation B2=c*B12. But that was for ECM, not sure if that applied to P-1 and P+1.
There is no difference because all these algorithms use smooth numbers. The difference is that we are requiring that p-1, p+1 or p+a are smooth (the absolute value of a in ECM is less than 2*sqrt(p) by Hasse's theorem).

2021-05-08, 17:10   #80
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

7,529 Posts

Quote:
 Originally Posted by alpertron Since the time to run the algorithm on Prime95 is proportional to B1 and B2, there is a balance, so in general you can find that with B2 = 30 B1 you get more throughput.
I agree there is a balance. I disagree with a blanket statement that the best balance is B2/B1 = 30.

The balance depends entirely on how efficient stage 1 is compared to stage 2. As you noted, GMP-ECM has a super-efficient stage 2 which implies B2/B1 should be much higher. Recent versions of prime95 have seen an increase in stage 2 efficiency.

Prime95 does a binary search looking for the optimal B2. Prime95 defines this as the point where the effort of increasing B2 further would be better spent on increasing B1 according to Mihai's smoothness probability calculator based on Dickman's function.

2021-05-08, 19:48   #81
pinhodecarlos

"Carlos Pinho"
Oct 2011
Milton Keynes, UK

3·17·97 Posts

Quote:
 Originally Posted by masser More? Longer tasks/higher bounds? Different range?
Yes, please. Same range but 10x more candidates.

2021-05-08, 21:03   #82
masser

Jul 2003

1,667 Posts

Quote:
 Originally Posted by pinhodecarlos Yes, please. Same range but 10x more candidates.
Here's a file with 8x more candidates; this covers the 1.7M range of Mersenne numbers without a known factor.
Attached Files
 wtd_cp2.txt (65.0 KB, 28 views)

2021-05-08, 21:43   #83
pinhodecarlos

"Carlos Pinho"
Oct 2011
Milton Keynes, UK

3·17·97 Posts

Quote:
 Originally Posted by masser Here's a file with 8x more candidates; this covers the 1.7M range of Mersenne numbers without a known factor.

Thank you, it is running.

 2021-05-08, 22:04 #84 pinhodecarlos     "Carlos Pinho" Oct 2011 Milton Keynes, UK 3·17·97 Posts Got unsupported assignment but they are running and also not uploading results to the server. Last fiddled with by pinhodecarlos on 2021-05-08 at 22:04
2021-05-08, 23:30   #85
alpertron

Aug 2002
Buenos Aires, Argentina

2·683 Posts

Quote:
 Originally Posted by Prime95 I agree there is a balance. I disagree with a blanket statement that the best balance is B2/B1 = 30. The balance depends entirely on how efficient stage 1 is compared to stage 2. As you noted, GMP-ECM has a super-efficient stage 2 which implies B2/B1 should be much higher. Recent versions of prime95 have seen an increase in stage 2 efficiency. Prime95 does a binary search looking for the optimal B2. Prime95 defines this as the point where the effort of increasing B2 further would be better spent on increasing B1 according to Mihai's smoothness probability calculator based on Dickman's function.
The number was based in the tests I performed in 2015. If in the meantime, you optimized step 2, it is clear that you can use a larger value of B2 / B1.

2021-05-08, 23:55   #86
masser

Jul 2003

1,667 Posts

Quote:
 Originally Posted by pinhodecarlos Got unsupported assignment but they are running and also not uploading results to the server.
Hmm. I'm not sure what to do here. I exclusively use the manual assignments uploader.

It looks like your results are being recorded here.

There don't appear to be blocking assignments (someone else has a reservation on the exponent) here.

Maybe it will soon be time for a Primenet P+1 reservation system?

 2021-05-09, 07:53 #87 pinhodecarlos     "Carlos Pinho" Oct 2011 Milton Keynes, UK 3·17·97 Posts Yep, they are now reporting correctly. Not sure why yesterday I had those messages but to be honest I didn’t bother to investigate since I was already in bed…lol My progress is 4/hour.
 2021-05-10, 17:07 #88 ATH Einyen     Dec 2003 Denmark 7·11·41 Posts Not sure how this "P+1" factor was found in stage1 with B1=550,000 ? 8676163 : P=538825884739595715103 B1 is larger than 5 of 6 factors of P-1 and larger than 4 of 6 factors of P+1. P+1=25 * 19 * 43 * 15973 * 568091 * 2271287 P-1= 2 *3 * 2953 * 12163 * 288181 * 8676163 Maybe stage2 ran but was not reported? Last fiddled with by ATH on 2021-05-10 at 17:09

 Similar Threads Thread Thread Starter Forum Replies Last Post Prime95 PrimeNet 103 2012-04-09 07:39 davieddy Lounge 34 2012-03-17 02:03 diamonddave GPU to 72 18 2011-12-06 19:56 Kees Puzzles 4 2006-04-07 07:17 Citrix Prime Sierpinski Project 5 2004-10-31 12:25

All times are UTC. The time now is 19:40.

Sun Jul 25 19:40:29 UTC 2021 up 2 days, 14:09, 0 users, load averages: 2.19, 1.94, 1.75