Strategic P-1 for DC (requires 24GB or more memory for prime95)

[kriesel]

+40:
Pfactor=1,2,105492967,-1,77,1
to
Pfactor=1,2,105662507,-1,77,1

[Andrew Usher]

Originally Posted by S485122

I think that the reservation posts are periodically cleaned (deleted). Other posts might be kept.

Doesn't anyone _know_, though? Posts being deleted seemingly at random or without reason is not trivial; if it were just reservation posts no one would care - they don't matter even to the poster - but it is not, and don't know what the criterion might be. The first post after that gap is one by James replying to (presumably) a non-reservation post that has been lost. There is at least one earlier gap between Jan. 24 and Mar. 8 that contained a post by me (not important anymore but not a reservation either), so it appears that long runs of posts are deleted to 'clean up' rather than any separation of reservation posts.

[kriesel]

+50:
Pfactor=1,2,105691171,-1,77,1
to
Pfactor=1,2,107274283,-1,77,1

[kriesel]

It appears to me that posts remain in this thread if, in the judgement of Prime95 (George), the post is sufficiently informative, such as

• contains directions (post 1),
• contains available work listings (post 2; frequently updated),
• end user posts asking or discussing how to proceed, that may be sufficiently useful and are not too wrong or irrelevant,
• communications from George or others about issues identified,
• work commitments that have not yet been entirely completed,
• work commitment posts Prime95 hasn't gotten around to reevaluating & perhaps removing after the commitments in the respective post are determined entirely complete (possibly currently posts 17-20; 33; 39-42; 45-47; 53; 57-61 are candidates for deletion; 62 and later are not yet).

There are higher priority uses for George's time than fastidious maintenance of a forum thread, so some lag or flexibility is a good thing.
And posts George hasn't seen yet would remain, unless they are so egregious that other moderators take action.
Work commitment posts soon after they're posted get modified in place when George updates the available work listing to reflect them.
IIRC when George is traveling and James Heinrich is filling in re this thread, he treads lightly.

[Reboot It]

Originally Posted by kriesel

but no assignment: 101916359 (Reboot It)
already assigned: 101915971 (Reboot It)
so +38 instead.
Perhaps someone with access to user email addresses could contact Reboot It to coordinate with this thread.

My apologies if I have unwittingly crossed over into P-1 work in this thread. I'm afraid I spend much less time on this forum than I used to... Thank you James for e-mailing me to bring it to my attention.

Whenever picking up P-1 work, I have always strictly checked that work is not already assigned to anyone, so I didn't know about this, and, to be honest, I had always assumed that the mersenne.org server was authoritative on the availability of work assignments.

I shall make a note in future either to avoid the range published in this thread, or to notify and claim any P-1 assignments in the list.

I think our goals are the same: to locate exponents with wildly inadequate historic P-1 and achieve thorough P-1 testing before they come up for double-checking.

[kriesel]

Originally Posted by Reboot It

My apologies if I have unwittingly crossed over into P-1 work in this thread. I'm afraid I spend much less time on this forum than I used to... Thank you James for e-mailing me to bring it to my attention.
...

I think our goals are the same: to locate exponents with wildly inadequate historic P-1 and achieve thorough P-1 testing before they come up for double-checking.

Thanks for responding, courteously and well. George's list in post 2 is an easy way to select targets for what I think of as remedial P-1. Posting those taken in this thread so the rest of us don't attempt them, and ensuring they get assigned (easily via manual communication with server update) will help things run smoothly.

I think our goals or the results with differing hardware may differ somewhat in the details. I intend total-compute-effort-optimal P-1 testing, and leave selection of bounds to prime95 via Pfactor= worktodo entries, and it typically optimizes for net expected compute time savings to large bounds on 17, 60 or more GiB ram allowed per worker, but not nearly as large as your bounds.

 17 GiB allowed example: M118277611 completed P-1, B1=687000, B2= 75182940
 60 GiB allowed example: M101990923 completed P-1, B1=555000, B2=181198290
250 GiB allowed example: M158914517 completed P-1, B1=856000, B2=568869840

Even at 1/4 TiB, I don't get nearly 1G for B2, nor 1M for B1 as a result of prime95's bounds optimization within allowed ram constraints.
How many TiB of ram per worker do you have set as allowed, on what hardware? Are you specifying bounds with Pminus1 worktodo entries or forcing high number of tests saved? https://www.mersenne.org/report_expo...1990923&full=1 shows B1 = 5M, B2 > 3.2G.
https://www.mersenne.ca/prob.php?exp...&b2=3205849500 shows 9.9% probability of factoring for 1036. GHD effort, vs. https://www.mersenne.ca/prob.php?exp...0&b2=181198290 6.86% at 87. GHD P-1 effort, so the same hardware would be able to do nearly 12. of the latter in place of one of the former. Since some of the first-PRP testers are skipping stage 2, doing many adequately is superior IMO to doing one to huge and perhaps uneconomic B1 and B2. While P-1 GHD and PRP GHD are no longer comparable in V30.8 or above, note that PRP on the same Mersenne number takes 406. GHD.

See also https://www.mersenneforum.org/showpo...0&postcount=15

[Reboot It]

Originally Posted by kriesel

I think our goals or the results with differing hardware may differ somewhat in the details.

That is more accurately put!

I am using Pminus1 in the worktodo, specifying manual B1 values. Prime95 (v30.8 for reference) assigns its own B2 value. I am fortunate to be able to allocate 80GiB to the P-1 work, hence the large B2 values.

I believe George has said in the past that one should "do what you find fun" when choosing work types, as long as it doesn't hinder or harm the project. I personally happen to find it far more satisfying to stretch each exponent to much higher bounds than the defaults, but I have only selected work with only one LL test completed and that has B1=B2, and B1 being the low default bound (such as was often used by CurtisC's machines in the past), so I am never redoing any Stage 2 work. Of course I do realise this is sub-optimal in terms of overall project throughput.

If anyone is ever interested, I have also been keeping the old P-1 savefiles, just in case someone wants to take them even further! (I'm up to about 1,000 files now.)

[kriesel]

+48:
Pfactor=1,2,107276237,-1,77,1
to
Pfactor=1,2,107506187,-1,78,1

[kriesel]

+47:
Pfactor=1,2,107506213,-1,78,1
to
Pfactor=1,2,107675257,-1,78,1

I already completed Pfactor=1,2,107276237,-1,77,1 in the previous batch

[kriesel]

+50:
Pfactor=1,2,107675417,-1,78,1
to
Pfactor=1,2,107758813,-1,78,1

[James Heinrich]

Just hijacking this thread slightly to say there are similar assignments available to re-do P-1 affected by the stage-2 bug in Prime95 v30.10-12.
Currently 411 assignments remaining:
https://www.mersenneforum.org/showthread.php?t=28707