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

[petrw1]

Originally Posted by ZacHFX

Jumping back in via GPU72, will snag a bunch of assignments to keep my house warm and the factors flowing.

[masser]

Is anyone working 8.6M? I'm eyeing that range as my next target after 14.0M is complete. Also, JIC anyone is curious/notices, my production rate on 8.6M will slow for the first few weeks while I use my desktop cpus on another deliverable.

[petrw1]

Originally Posted by masser

Is anyone working 8.6M? I'm eyeing that range as my next target after 14.0M is complete. Also, JIC anyone is curious/notices, my production rate on 8.6M will slow for the first few weeks while I use my desktop cpus on another deliverable.

You like the challenges don't you?
That's great; someone has to do them.

I don't see any evidence of recent work there other than TF which is now complete.
One more TF bit is scheduled but that could be 3 or 4 months from now.
It should remove about 20 factors.

If I may offer some calculations .. use them as you see fit, or not at all: it's your work in the end.
Based on the current Bounds and Calculated success rates compared to possible new Bounds and Estimated success rates:

A new success rate of 7% should get you about 115 factors; 8% about 139 factors.

%:  B1          B2              GhzDays
7%: 4,137,402	144,809,070	4.468210
8%: 6,813,150	272,526,000	8.060600

Thanks ...

[masser]

Originally Posted by petrw1

I don't see any evidence of recent work there other than TF which is now complete.
One more TF bit is scheduled but that could be 3 or 4 months from now.
It should remove about 20 factors.

A new success rate of 7% should get you about 115 factors; 8% about 139 factors.


Thanks ...

Great! I will take 8.6M then. When my P-1 machines become available, I will most likely target bounds near that 8% probability with 8GhzD effort setting.

[lisanderke]

I've been catching up with this thread and I'd love to help out the 8.xM ranges with P-1 if any exponents ranges are available/worthwhile doing P-1 on. I've currently got about three days worth of P-1 work queued up in the 8.2M-8.3M range, re-doing P-1 from https://www.mersenne.ca/pm1_worst.php
I just noticed now that the 8.3M range is already below 2000 unfactored, but these aren't big workloads so I'll run these aswell :)


Is there an easy tool (besides the worst P-1 tool on mersenne.ca) with which I could easily add these exponents to my worktodo?

[petrw1]

Originally Posted by lisanderke

I've been catching up with this thread and I'd love to help out the 8.xM ranges with P-1 if any exponents ranges are available/worthwhile doing P-1 on. I've currently got about three days worth of P-1 work queued up in the 8.2M-8.3M range, re-doing P-1 from https://www.mersenne.ca/pm1_worst.php
I just noticed now that the 8.3M range is already below 2000 unfactored, but these aren't big workloads so I'll run these aswell :)

Is there an easy tool (besides the worst P-1 tool on mersenne.ca) with which I could easily add these exponents to my worktodo?

Welcome...
The tool you mentioned will easily give you P-1 assignments for your worktodo.
My concern (selfishly for the sake of the goals of this project) is that the Bounds (B1&B2) that are recommends may not be high enough to find the number of factors required.

That said, if you stick with the ranges with the lower number of factors remaining they should be fine.
For example:
8.2M
8.9M
9.4M

[lisanderke]

Oh, I hadn't considered that. Currently Prime95 is reporting a 7.5% chance of finding factors for 8.5M exponents. Would you be able to provide me with a replacement primality_tests_saved number for the 8.5M range? I'd be happy to work on the other ranges you've provided with the bounds/tests_saved the tool provides, but I'll be sure to post in this thread when I'm ready to do so.
Currently the assignments are of this form:
Pfactor=F2----------------------------E5,1,2,8569189,-1,64,6



(PS: Someone else has been turning in P-1 assignments that were obtained with the "worst P-1" tool after I had reported success on them. I assume they didn't notice their assignments weren't receiving AIDs, or they were doing P-1 factoring without having them assigned to primenet automatically on some other program. Currently I've moved on from, and unreserved the 8.2M - 8.3M range I was working on.)

[petrw1]

Originally Posted by lisanderke

Oh, I hadn't considered that. Currently Prime95 is reporting a 7.5% chance of finding factors for 8.5M exponents. Would you be able to provide me with a replacement primality_tests_saved number for the 8.5M range? I'd be happy to work on the other ranges you've provided with the bounds/tests_saved the tool provides, but I'll be sure to post in this thread when I'm ready to do so.
Currently the assignments are of this form:
Pfactor=F2----------------------------E5,1,2,8569189,-1,64,6

Detailed explanation is below ... if you have no interest in that here is the short answer:
The 7.5% is misleading.
I don't know why it is using '64' as the bits factored when these exponents have been factored up to 72 bits.
If you were to change the 64 to a 72 your percent would drop from 7.5% to about 3.4%.

To get 59 more factors out of 2058 exponents you need a success rate that is about 3% higher for the entire range (all 2058 exponents).
Based on 72 bits of TF your exponent has a current success rate of about 1.766%.

Change the 64 to 72 and verify that tests_saved=6 results in about 3.4%.
Then make the test_saved high enough that the resultant success rate is at least 5% (3%+ above the current value)
Once you know the desired value for tests_saved generate your desired pool of assignments then use your favorite editor to change the worktodo accordingly.

Details:
What works best for me is I start here and by sorting by B1 and/or B2 and eyeballing the entire range I get a sense for the current ranges of B1/B2 values.
Here you can determine what the current success rates are.
For example, your exponent has B1=100000, B2=1825000 and is factored to 72 bits and has a current success rate of 1.766%.
You obviously cannot check all of them; pick one with a common current B1/B2.
Note how many more are similar. In this case, about 2/3 of the exponents in this range have similar B1/B2.

You will likely find that for most ranges there will be a portion of the exponents that have adequately large B1/B2 that further P-1 is not worthwhile.
This will reduce the pool of exponents you can use. This, in turn, will increase the required success rate for them.
Once you have analyzed the range you can determine what exponents have adequate success rates for a rerun; and the required B1/B2 values to get the success rates you need.
Then use those values to generate Pfactor or Pminus1 assignments as desired.

(PS: Someone else has been turning in P-1 assignments that were obtained with the "worst P-1" tool after I had reported success on them. I assume they didn't notice their assignments weren't receiving AIDs, or they were doing P-1 factoring without having them assigned to primenet automatically on some other program. Currently I've moved on from, and unreserved the 8.2M - 8.3M range I was working on.)

That's too bad when that happens.

Thanks again
Wayne

[petrw1]

All other ranges above 49.6M and below 108.0M are complete.
All other 4x.xM ranges are complete.

49.6M only needs 3 more factors.
Anton Repko has been methodically running TF and some PP1.
He is a few days from completing TF to 76 bits.
I'm hoping for more factors from him and with luck completing it.

Anton: if your intention is to finish 49.6M yourself that would be wonderful and I'll stay out of your way.
Just let me know here. Otherwise, I've indicated below my opinion of what needs to happen next.

=======================================

If Anton does not finish it I suggest a little more aggressive P-1 should take care of it.
There are 388 exponents that could have more P-1 done; those with B1/B2 currently up to: 890000 & 22000000.

Statistically running P-1 on these remaining exponents with B1/B2 of 2M/60M (or something comparable)
is expected to find 6 factors; we'll need at most 3 so I am optimistic this is adequate.

I welcome a volunteer; I'm even willing to produce the PMinus1 statements if so desired.

My current work to complete the 4 remaining 3x.xM ranges will take about 3 more weeks.

If no one takes up the challenge I will take 49.6M then.

Thanks
Wayne

[Zhangrc]

Originally Posted by petrw1

All other ranges above 49.6M and below 107.2M are complete.

107.2M is also complete (1991 unfactored).
Please focus on 108.3M which still has 96 factors to go. (I only TFed to find 3 factors, ETA is 2 years, would anyone like to help?)

[9vLYvxWUJ7Dy2Hb]

Given that I'm only making minimal progress on this range and someone else (Sid & Andy) is having way more throughput I'll limit myself to everything below 31219961


This means the following range is up for grabs again: 31220003 to 31249993