COMPLETE!!!! Thinking out loud about getting under 20M unfactored exponents - Page 87

Zhangrc · 2021-11-27, 01:24

Quote:

Originally Posted by chalsall

I've added a column to the To Below Two Thousand Unfactored Candidates report: "Avg B2".

It should be "exp(avg ln B2)", since the chance of finding a factor in P-1 grows almost logarithmically with B2 (assuming B2=k*B1 where k is a constant).

petrw1 · 2021-11-27, 04:05

They are only TF'd to 73 bits but each range only needs 19 more factors.

If you choose to TF them to 74 they will likely clear.
If you choose to P-1 them with modest bounds they will also clear.

Luminescence · 2021-11-27, 04:53

Quote:

Originally Posted by petrw1

They are only TF'd to 73 bits but each range only needs 19 more factors.

If you choose to TF them to 74 they will likely clear.
If you choose to P-1 them with modest bounds they will also clear.

Also taking 20.3M

lisanderke · 2021-11-28, 23:26

I'll be doing P-1 in the 10.xM ranges (10.4M-11.0M) with B1=700k to 800k and B2: whatever v30.8b2 assigns :)

petrw1 · 2021-11-29, 19:44

That leaves the highest undone in our ranges of interest at 29.8.

Wow Thanks everyone.

nordi · 2021-11-29, 21:33

Quote:

Originally Posted by kruoli

While stage 2 in worker #1 was running (using 110-115 % of the memory I had allowed it)

How much did you allow? I set mine to 20GB for your first task and that limit was kept.

kruoli · 2021-11-29, 21:38

IIRC I set it to 20 GB and had 23 GB usage. On the same machine I regularly had overallocation with ECM. George got a lot of them sorted out.

masser · 2021-11-29, 21:59

Quote:

Originally Posted by petrw1

That leaves the highest undone in our ranges of interest at 29.8.

Wow Thanks everyone.

SethTro · 2021-11-30, 01:59

I'm re-running stage 2 with 30.8v2 and founding a few extra factors from larger B2

https://www.mersenne.org/report_expo...6907619&full=1

Stage 2 is several times faster. From memory it was taking 5-10K core seconds to complete B2=110M vs 2K core seconds to complete B2=414M now.

Luminescence · 2021-12-01, 02:19

Quote:

Originally Posted by firejuggler

If this is of any help, here are the factor I found 6.2-6.3M with their B1/B2 value

I tried those exponents and found another one on 6228493. I found a composite one with your factor and another larger one at B1=3,000,000 and B2=13,938,874,950

petrw1 · 2021-12-01, 03:11

George in 30.8 is working on determining the best B1/B2 for this new version.
Because Stage 2 is several times faster than it used to be the optimal B2 is a much higher multiple of B1.
This will equate to a higher success rate and more factors found on average.
Most of the time P-1 is run on fresh exponents with no prior P-1.

However, for this sub-project most of the P-1 work is on exponents that have already had P-1 done; just to relatively low bounds.
With newer, faster hardware it is reasonable to re-do P-1 to higher bounds and factor more exponents.
Along with that, the goal of this sub-project is to find a defined quantity of factors as efficiently as possible.
So in the past I've done a lot of analysis and some trial-and-error and *have* a pretty good handle of the recommended bounds for each remaining range.

However, with version 30.8 I now *had* a pretty good handle.

The basic formula remains the same:
- Note how many factors are required.
- Analyze the current average P-1 success rate.
- Calculate the new success rate that is required to produce the required number of factors.
- Determine the new P-1 bounds that achieve that success rate.

For example, prior to 30.8 if I needed a +3% success rate my new Bounds would be in the 1.5M/45M range.
A 30x B2/B1 ratio was reasonable in these versions.
For an exponent in the 28M these bound give a 4.58% success rate....here
I'm assuming the current success rate is about 1.58%; reasonable.

With the new version recommending about a 200x B2/B1 ratio to get the same 4.58% success rate would require bounds of about 530K/106M ... here

That may not seem like a problem until you consider that the current B1 for most of the exponents is over 530K (those that are lower are not much lower).
Therefore, bounds such as these make it unlikely that a Stage 1 factor will be found.
But maybe that is not a big problem with Stage 2 being so much faster.

For now I'm just not quite sure what to suggest.
- Use as above and accept Stage 1 is unlikely to find a factor.
- Increase B1/Reduce B2 to 100x to make Stage 1 more productive but lose some overall P-1 efficiencies.
- Increase B1/Keep B2 at 200x; find more factors sooner but with more total effort .... but in less time due to the speed of 30.8
- Something else?

Opinions?
Thanks

P.S. In the same clock time Stage 2 has 3 to 4 times the odds of finding a factor of Stage 1. Seems Stage 1 is not as significant/effective in 30.8

2021-11-27, 04:05	#948
petrw1 1976 Toyota Corona years forever! "Wayne" Nov 2006 Saskatchewan, Canada 5,179 Posts	I've added 20.3 and 20.5 to the ToDo list on Post #1 They are only TF'd to 73 bits but each range only needs 19 more factors. If you choose to TF them to 74 they will likely clear. If you choose to P-1 them with modest bounds they will also clear.

2021-11-28, 23:26	#950
lisanderke "Lisander Viaene" Oct 2020 Belgium 1101101₂ Posts	I'll be doing P-1 in the 10.xM ranges (10.4M-11.0M) with B1=700k to 800k and B2: whatever v30.8b2 assigns :) Last fiddled with by lisanderke on 2021-11-28 at 23:27

2021-11-29, 19:44	#951
petrw1 1976 Toyota Corona years forever! "Wayne" Nov 2006 Saskatchewan, Canada 5,179 Posts	Hurrah 49.6 is cleared!!!! That leaves the highest undone in our ranges of interest at 29.8. Wow Thanks everyone.

2021-12-01, 03:11	#957
petrw1 1976 Toyota Corona years forever! "Wayne" Nov 2006 Saskatchewan, Canada 5,179 Posts	Pondering 30.8 with respect to this project. George in 30.8 is working on determining the best B1/B2 for this new version. Because Stage 2 is several times faster than it used to be the optimal B2 is a much higher multiple of B1. This will equate to a higher success rate and more factors found on average. Most of the time P-1 is run on fresh exponents with no prior P-1. However, for this sub-project most of the P-1 work is on exponents that have already had P-1 done; just to relatively low bounds. With newer, faster hardware it is reasonable to re-do P-1 to higher bounds and factor more exponents. Along with that, the goal of this sub-project is to find a defined quantity of factors as efficiently as possible. So in the past I've done a lot of analysis and some trial-and-error and have a pretty good handle of the recommended bounds for each remaining range. However, with version 30.8 I now had a pretty good handle. The basic formula remains the same: - Note how many factors are required. - Analyze the current average P-1 success rate. - Calculate the new success rate that is required to produce the required number of factors. - Determine the new P-1 bounds that achieve that success rate. For example, prior to 30.8 if I needed a +3% success rate my new Bounds would be in the 1.5M/45M range. A 30x B2/B1 ratio was reasonable in these versions. For an exponent in the 28M these bound give a 4.58% success rate....here I'm assuming the current success rate is about 1.58%; reasonable. With the new version recommending about a 200x B2/B1 ratio to get the same 4.58% success rate would require bounds of about 530K/106M ... here That may not seem like a problem until you consider that the current B1 for most of the exponents is over 530K (those that are lower are not much lower). Therefore, bounds such as these make it unlikely that a Stage 1 factor will be found. But maybe that is not a big problem with Stage 2 being so much faster. For now I'm just not quite sure what to suggest. - Use as above and accept Stage 1 is unlikely to find a factor. - Increase B1/Reduce B2 to 100x to make Stage 1 more productive but lose some overall P-1 efficiencies. - Increase B1/Keep B2 at 200x; find more factors sooner but with more total effort .... but in less time due to the speed of 30.8 - Something else? Opinions? Thanks P.S. In the same clock time Stage 2 has 3 to 4 times the odds of finding a factor of Stage 1. Seems Stage 1 is not as significant/effective in 30.8 Last fiddled with by petrw1 on 2021-12-01 at 05:24 Reason: P.S.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Thinking of Joining GPU to 72	jschwar313	GPU to 72	3	2016-01-31 00:50
Thinking about lasieve5	Batalov	Factoring	6	2011-12-27 22:40
Thinking about buying a panda	jasong	jasong	1	2008-11-11 09:43
Loud thinking on irregular primes	devarajkandadai	Math	4	2007-07-25 03:01
Question on unfactored numbers...	WraithX	GMP-ECM	1	2006-03-19 22:16

2021-11-29, 21:38	#953
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 1111001101₂ Posts	IIRC I set it to 20 GB and had 23 GB usage. On the same machine I regularly had overallocation with ECM. George got a lot of them sorted out.

2021-11-30, 01:59	#955
SethTro "Seth" Apr 2019 2⁴×3³ Posts	I'm re-running stage 2 with 30.8v2 and founding a few extra factors from larger B2 https://www.mersenne.org/report_expo...6907619&full=1 Stage 2 is several times faster. From memory it was taking 5-10K core seconds to complete B2=110M vs 2K core seconds to complete B2=414M now.