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

[axn]

I am coming up to the end of 5.9x fast (although it wouldn't look like it from stats). I need another range to start work on -- preferably a hard one. Is there a recommended / most wanted range?

[petrw1]

Originally Posted by axn

I am coming up to the end of 5.9x fast (although it wouldn't look like it from stats). I need another range to start work on -- preferably a hard one. Is there a recommended / most wanted range?

If you'd like a hard range on the lower end there is 10.4 or 10.7.
Anonymous may be there in a couple months with one more bit of TF.
Or you can claim one of these range and Chris will release it from TF and it will be all yours.

On the higher end we have ranges you will see in the first post of this thread.
Though, the only unclaimed are:
27.0 ... probably the toughest remaining
22.3
25.3 ... though Anton was working there as recently as 10 days ago.
21.9 will complete TF in about a week
20.8 will be a tough one too but it needs at least 1 more bit of factoring first ... unless you have a decent GPU for that too.

Thanks

[petrw1]

I was patiently waiting for 3xM and 4xM to clear.
This is a major milestone with all these clearing in November.

52 more ranges cleared!!! by far the biggest jump to date.!!!!
5.2, 5.4, 5.5, 6.2, 7.4, 9.2, 9.4,
11.3, 13.0, 14.0, 15.4, 17.3, 18.1, 18.5, 19.2,
20.3, 20.9, 21.3, 21.4, 22.0, 22,4, 23.2, 23.4, 23.9,
24.0, 24.1, 24.2, 24.4, 24.8, 24.9, 25.4, 25.7, 25.9,
26.3, 26.6, 27.2, 27.3, 27.5, 27.7, 28.1, 28.2, 28.5, 28.9, 29.0, 29.7,
30.5, 30.8, 31.2, 32.7, 34.4, 35.3, 49.8

TOTALS to date:
374 total ranges cleared or 75.25% (123 ranges remaining)
11 with less than 20 to go.
114 with less than 100 to go.
3,993 more factored (48,098)....87.09% total factored.

Category	Start	Today
Lowest ToDo	1.8	5.9	(The lowest range still over 1999)
Highest ToDo	86.3	29.8	(The highest remaining for this project ... all higher will be done via standard Prime95 protocols)
Low UnDone	86.4	108.0	(The lowest still over 1999 that Prime95 will eventually clear on its own)
High UnDone	999.8	999.8	(The highest remaining for Prime95)

Current Activities
I have all my P1 firepower in the 2xM range.
There are 24 ranges left there....several of you are helping there too either with TF or P-1.
It has 7 more ranges needing TF74 (My GPUs and several others are there); and then a half dozen or so will benefit from TF75.
That is, if TF can keep up now that P-1 is getting so much faster .... but that's another story.
My crystal ball has 2xM completing in the next 6 or 8 months.

On the TF side, sadly we have lost SRBase and most of Judger (at least for now); thankfully Anonymous is still chugging.
That said, I think TF is in good shape now dramatically reducing the remaining P-1 effort.

Many of you are P1'ing and TF'ing on all the remaining lower ranges.

Where is the help most required?
We are on the home stretch now; I see light at the end of the tunnel (and I don't think it's an oncoming train)
With all the attention this project is getting now I think most of you know exactly where we are at and are helping appropriately....thanks a lot.

You may have seen a post a months or so ago indicating that I have a chart on post 1 of this thread for the remaining 2xM ranges; where they are at and who has claimed them.

I hope to expand this table soon to all active ranges below as well.

Thanks!!! THANKS!!!!!!

[axn]

Originally Posted by petrw1

On the higher end we have ranges you will see in the first post of this thread.
Though, the only unclaimed are:
27.0 ... probably the toughest remaining

Alright. In that case, I would like to work on 27.0. I think a prob of 5.6% should do the trick which means something like 600K / 400M should be sufficient. I might tweak the bounds once I start some actual tests.

[alpertron]

Originally Posted by alpertron

I'm running a more intensive test. The contents of the file worktodo.txt is:

[Worker #1]
Pminus1=1,2,9325159,-1,50000000,50000000000,"8056937377,16207126343,95788033249,799651034569,2894827758689,1416068886691577,36098431074477463,635315498297290933609,6214794856879965540791"

It took 26 hours for step 1 in an I5 3470. and now it is running step 2 for 35 hours with 8 GB used for Prime95. According to the values above, probably Prime95 will require 10 days to complete step 2.

Prime95 has just finished this task with no hang:

processing: P-1 no-factor for M9325159 (B1=50,000,000, B2=50,001,265,860)
CPU credit is 1312.7590 GHz-days.

[petrw1]

Originally Posted by axn

Alright. In that case, I would like to work on 27.0. I think a prob of 5.6% should do the trick which means something like 600K / 400M should be sufficient. I might tweak the bounds once I start some actual tests.

According to my calculations that percentage should give you 78 factors....well done.
I like your plan ... I'd start with that for the current B1=B2 and hope for about half the factors.
Then depending on your luck adjust if necessary.

[techn1ciaN]

I'm just watching this project from the sidelines, so my apologies if this question is obvious or has already been asked: Does the assumption that "all higher will be done via standard Prime95 protocols" partially rely upon P-1? If yes, has the reality of bad P-1 been accounted for? Example: An exponent without P-1 done is assigned for PRP to a user who has not allocated Prime95 enough memory for stage 2. The user turns in P-1 with B2=B1, and then the exponent is retired from further assignment once they turn in the PRP. Doesn't this make it possible that even if a 100k exponent range would theoretically get to <2,000 unfactored without further intervention, it might not actually — leaving a (smaller, but) lasting role for this group in "manually" cleaning up bad P-1?

Along similar lines, the P-1 tests_saved parameter in P-1 and FTC PRP assignment lines will eventually change from 2 to 1 (hopefully sooner rather than later for the sake of overall GIMPS throughput). Has this been accounted for?

[petrw1]

Originally Posted by techn1ciaN

I'm just watching this project from the sidelines, so my apologies if this question is obvious or has already been asked: Does the assumption that "all higher will be done via standard Prime95 protocols" partially rely upon P-1? If yes, has the reality of bad P-1 been accounted for? Example: An exponent without P-1 done is assigned for PRP to a user who has not allocated Prime95 enough memory for stage 2. The user turns in P-1 with B2=B1, and then the exponent is retired from further assignment once they turn in the PRP. Doesn't this make it possible that even if a 100k exponent range would theoretically get to <2,000 unfactored without further intervention, it might not actually — leaving a (smaller, but) lasting role for this group in "manually" cleaning up bad P-1?

Along similar lines, the P-1 tests_saved parameter in P-1 and FTC PRP assignment lines will eventually change from 2 to 1 (hopefully sooner rather than later for the sake of overall GIMPS throughput). Has this been accounted for?

Good questions.
1. Correct .... we are at the mercy of poor P-1. However as the required TF bit level increases this will matter less.
So far every range from 86.4 to 107.9 has cleared. As noted earlier 108.3 looks iffy.
So I accept there may be a bit of cleanup required later.
Two years ago that was so far off I didn't give it a thought.
However, with all the extra help here lately ... and the much faster P-1 we could complete all lower ranges by the end of 2022.
If that is the case then by then there will be very little cleanup, if any. And I'm pretty sure it will show up even slower than I alone could manage.

2. With P-1 so much faster in 30.8 it might make sense to let it go deeper and leave it at 2 tests saved ... or even more. This is similar to what happened when GPUs appeared for TF. Several bit levels were added to TF before LL (or PRP) were done.

[masser]

Originally Posted by techn1ciaN

Does the assumption that "all higher will be done via standard Prime95 protocols" partially rely upon P-1? If yes, has the reality of bad P-1 been accounted for? Example: An exponent without P-1 done is assigned for PRP to a user who has not allocated Prime95 enough memory for stage 2. The user turns in P-1 with B2=B1, and then the exponent is retired from further assignment once they turn in the PRP. Doesn't this make it possible that even if a 100k exponent range would theoretically get to <2,000 unfactored without further intervention, it might not actually — leaving a (smaller, but) lasting role for this group in "manually" cleaning up bad P-1?

Primenet trial factoring will take care of almost all of the remaining high (>100M) ranges. The few stragglers that survive trial factoring won't survive the P-1 stage, even if a fair amount are poorly P-1'd.

[SethTro]

@techn1ciaN

My opinion without running any queries or writing code is no we don't need to do anything special. We should keep querying the counts in ranges up to 120M (where TF increases to 73 bits), but 99.9% of them will finish even if we perform sub-par P-1. There's only 119 ranges between 108.0 and 119.9 so maybe we'll have to do a little additional TF for one or two of those ranges (e.g. 108.3) but it's likely people like Zhangrc will notice before the wavefront gets there.

[chalsall]

Originally Posted by petrw1

As noted earlier 108.3 looks iffy.

OK... Since this seems to be a concern, I have brought 108.3M (actually reserved from Primenet, and those candidates not already assigned) into GPU72 to TF to 77 bits, and then P-1 (usually extremely well).