k=13236795 - Page 10

bur · 2021-12-02, 17:28

How to proceed (sieve, LLR, P-1) depends on which method removes candidates faster. For sieving that's a bit hard to determine since it removes an n=10M almost as fast as n=1M while the LLR of the former takes 100 times more.

For P-1 vs LLR it's easy though. Just time how long LLR takes to remove 10 candidates and how long P-1 would take in the same range. I strongly suspect P-1 doesn't make sense. It's only useful for Mersennes because they are so large and have those specially formed factors including the exponent.

Here's how I do it: do a few LLR tests and write down the average time. Then sieve for 1-2 days and see if that removes candidates faster or not. If yes, continue sieving, if not, continue LLR until an LLR test again takes longer to remove a candídate than sieving did.

ValerieVonck · 2021-12-02, 18:25

Quote:

Originally Posted by bur

How to proceed (sieve, LLR, P-1) depends on which method removes candidates faster. For sieving that's a bit hard to determine since it removes an n=10M almost as fast as n=1M while the LLR of the former takes 100 times more.

For P-1 vs LLR it's easy though. Just time how long LLR takes to remove 10 candidates and how long P-1 would take in the same range. I strongly suspect P-1 doesn't make sense. It's only useful for Mersennes because they are so large and have those specially formed factors including the exponent.

Here's how I do it: do a few LLR tests and write down the average time. Then sieve for 1-2 days and see if that removes candidates faster or not. If yes, continue sieving, if not, continue LLR until an LLR test again takes longer to remove a candídate than sieving did.

Thank you for your insights
1. I have retested the primes found by gd_barnes : they are confirmed
2. On my i3 I am sieving the input from gd_barnes to 260T: eta 03/22
3. On my i5 imac I am llr’ing the range 1.3m to 1.4m each candidate takes 733 seconds
4. Sieving on that latest machine I am eliminating a candidate each 500 - 550 seconds
4.1. On a sidenote gd_barnes Saïd to me you are missing factors due to a wrong setting in NewPgen, my file ranges from 1.3m to 50m and I already sieved to 260T … taking this up from the beginning will take also take time
5. I will take a stab at p-1 when my current range is done or will try it on my i7 portable (laptop)

Ftm I am recovering from a mild Covid infection

Kind regards,
Valerie

VBCurtis · 2021-12-03, 01:51

If you're going to LLR test the entire file, then one should sieve until the factor-removal rate is half as fast (twice the seconds per factor) as an LLR test takes for the smallest candidate in the file.

This is due to the way the sieve scales- when you take candidates out of the sieve for LLR, the sieve speed does not change linearly. Since there is not much speed to be gained by taking candidates out, we leave the small candidates in longer than common-sense might suggest.

A rough scaling example: If you take out 10% of the candidates from the sieve file, the sieve only runs ~5% faster ("faster" as measured by p/sec).

ValerieVonck · 2021-12-08, 05:20

I gathered a few statistics:

- Sieving yields a factor each 552 seconds
- An llr test takes 735 seconds for the range 1.3M -> 1.4M (which I am currently busy testing)
- A P-1 test takes 30 minutes, with bounds B1 = 1M
- An ECM test takes 30 minutes for each curve, with bounds B1 = 1M

For the moment, I will complete the range 1.3M -> 1.4M

Kind regards,
Valerie

Happy5214 · 2022-01-23, 09:28

P-1 is useless for Riesel candidates for the reasons bur listed. In fact, your tested B1 of 1M is higher than what GIMPS recommends even for wavefront (i.e. p~105M) Mersenne candidates (which is in the 800k range on my computer). Bounds that high are only used on numbers in the 1.3M-bit range when we're trying to (partially) factor it, not pretest before running a primality test.

ValerieVonck · 2022-05-29, 06:58

Range 1.3M -> 1.4M completed, no new primes to report.
All ranges re-sieved up unto 260T & LLR'ed to 1.3M, the outputs are matching my results
Will sieve until 300T

Kind regards,
Valerie

ValerieVonck · 2022-07-03, 15:32

With the energy prices surging in Europe, I have decided to halt my efforts on this project.
I have sieved the number to 270T en tested it to 1.4M.

Please find attached the link to my Google Drive where you can pickup this file.
If this does not work please PM your (Project responsible) with your mail address and I will then transfer the input to you.

It was a fun, and certainly a long and a happy time.

I hope I will back, when the prices are stabilizing.

https://drive.google.com/file/d/1LMp...ew?usp=sharing

Kind regards
Valerie

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
K=13236795 Status Post	ValerieVonck	15k Search	33	2005-10-31 22:54

2021-12-02, 17:28	#100
bur Aug 2020 796581e-4;32539e-3 2×5×73 Posts	How to proceed (sieve, LLR, P-1) depends on which method removes candidates faster. For sieving that's a bit hard to determine since it removes an n=10M almost as fast as n=1M while the LLR of the former takes 100 times more. For P-1 vs LLR it's easy though. Just time how long LLR takes to remove 10 candidates and how long P-1 would take in the same range. I strongly suspect P-1 doesn't make sense. It's only useful for Mersennes because they are so large and have those specially formed factors including the exponent. Here's how I do it: do a few LLR tests and write down the average time. Then sieve for 1-2 days and see if that removes candidates faster or not. If yes, continue sieving, if not, continue LLR until an LLR test again takes longer to remove a candídate than sieving did.

2021-12-03, 01:51	#102
VBCurtis "Curtis" Feb 2005 Riverside, CA 13263₈ Posts	If you're going to LLR test the entire file, then one should sieve until the factor-removal rate is half as fast (twice the seconds per factor) as an LLR test takes for the smallest candidate in the file. This is due to the way the sieve scales- when you take candidates out of the sieve for LLR, the sieve speed does not change linearly. Since there is not much speed to be gained by taking candidates out, we leave the small candidates in longer than common-sense might suggest. A rough scaling example: If you take out 10% of the candidates from the sieve file, the sieve only runs ~5% faster ("faster" as measured by p/sec).

2021-12-08, 05:20	#103
ValerieVonck Mar 2004 Belgium 7·11² Posts	I gathered a few statistics: - Sieving yields a factor each 552 seconds - An llr test takes 735 seconds for the range 1.3M -> 1.4M (which I am currently busy testing) - A P-1 test takes 30 minutes, with bounds B1 = 1M - An ECM test takes 30 minutes for each curve, with bounds B1 = 1M For the moment, I will complete the range 1.3M -> 1.4M Kind regards, Valerie

2022-01-23, 09:28	#104
Happy5214 "Alexander" Nov 2008 The Alamo City 2²×5×7² Posts	P-1 is useless for Riesel candidates for the reasons bur listed. In fact, your tested B1 of 1M is higher than what GIMPS recommends even for wavefront (i.e. p~105M) Mersenne candidates (which is in the 800k range on my computer). Bounds that high are only used on numbers in the 1.3M-bit range when we're trying to (partially) factor it, not pretest before running a primality test.

2022-07-03, 15:32	#106
ValerieVonck Mar 2004 Belgium 1101001111₂ Posts	With the energy prices surging in Europe, I have decided to halt my efforts on this project. I have sieved the number to 270T en tested it to 1.4M. Please find attached the link to my Google Drive where you can pickup this file. If this does not work please PM your (Project responsible) with your mail address and I will then transfer the input to you. It was a fun, and certainly a long and a happy time. I hope I will back, when the prices are stabilizing. https://drive.google.com/file/d/1LMp...ew?usp=sharing Kind regards Valerie