Trial Factor and P-1 assignments squatting in Cat 0 and Cat 1 ranges

[Uncwilly]

User Ryan Propper has 430 PM-1 assignments between 101M and 102M, all of which have yet to have a first time test.
They are 2 months old and should have expired as Cat 0 assignments.
There are 127 other P-1's assigned in the range most of them >30 days old and no progress.

[slandrum]

It looks like someone released all the Ryan Propper assignments, 429 assignments are now open below 102M

[Uncwilly]

And now another user has been picking up dozens of P-1 assignments. And other users have picked up P-1 assignments in the Cat 0 range.

[linament]

And now another user has been picking up dozens of P-1 assignments. And other users have picked up P-1 assignments in the Cat 0 range.

I guess I don't understand the issue. As far as I know, P-1 is the step prior to running a PRP test or a LL test. My quick sampling shows that P-1 has not been run on these exponents.

[Prime95]

It's not an issue as long as the users do their P-1 in a timely manner. In fact it is beneficial in that dedicated P-1ers usually dedicate more memory for stage 2 resulting in more factors found.

A year ago the dedicated P-1ers were keeping ahead of the LL/PRP wavefront. Ben Delo's huge contribution as well as gpuowl's improvements have made ir impossible for P-1ers to keep up with the wavefront.

Edit: I think the gripe is that P-1 assignments are not subjected to the same expiration rules as LL/PRP assignments.

[ATH]

Originally Posted by Prime95

Edit: I think the gripe is that P-1 assignments are not subjected to the same expiration rules as LL/PRP assignments.

YES! Also do we want P-1 assignment in Cat 0+1 ? It is already too late doing P-1 at that point? Maybe let Prime95/mprime do automatic P-1 test before PRP test like in the old days before LL tests.
Is P-1 really that important now when we only have to do 1 test + CERT?

In my opinion P-1 and TF assignments should be blocked in Cat 0 at least, and probably in Cat 1.


Now most of those 300+ P-1 assignments has just moved to Kriesel and Madpoo. Are they going to complete them in a timely manner? It would be better to release most of them to the general public and just keep a smaller number?

[kriesel]

Some here seem concerned about P-1 assignments in 101-102M. Don't worry. It's not a problem. It's a solution in progress, at least in regard to my portion of them.
Hundreds of exponents that needed P-1 were released from months-old stagnant assignments.
They still need P-1 performed. Expediting that so they can get justified PRP quickly too, or eliminated from further testing by finding a factor, is a good thing.

The hundreds in 101M-102M assigned for P-1 to Kriesel are prioritized on and distributed to multiple Cat0-rated cpus, multiple Radeon VII gpus, and other hardware. Should take only about 2 days from assignment to complete those now so queued. The cpus range mostly from 16GB to 128 GB ram and apply whatever bounds prime95's time savings optimization algorithm selects.
FYI a P-1 with 1M B1, 30M B2 on ~102M exponent (gpuowl 6.11 default bounds) is ~47 minutes of a Radeon VII, so they crank out ~30 P-1 completions/day/RadeonVII to more-than-sufficient B1 and B2.
If more 101M P-1 assignments are available later, more will be reserved and swiftly run. It takes a lot of such assignments to keep a Radeon VII busy.
That will speed reaching the 102M milestone, not delay it as some seem to express concern of.
Ben Delo is the main consumer of P-1 wavefront completions I produce.
When I produce less in the P-1 wavefront, his P-1 activity goes up noticeably in the top producer listing (and that subtracts slightly from his PRP throughput).

I've already submitted over 40 P-1 factoring results for 101M-102M for the past 12 hours' running. Including two factored and eliminated.
The rate over the next 24 hours will be higher, since some other work already in progress (including some of the pending 102-103M P-1) has completed. The PrimeNet connected cpus' results are submitted as completed. The gpus' results are manually reported in daily bursts.

The 30-day ETAs of MadPoo's P-1 assignments in this range are unfortunate, if those long ETAs are accurate.
Hopefully they will also be done to proper bounds that both retire the P-1 assignment task and provide near optimal total computation time savings.

Some primality tests now in progress in 101M-102M have inadequate P-1 (stage 1 only, or low bounds), or none performed yet, yet have already progressed to PRP assignments.
https://www.mersenne.org/report_expo...1066557&full=1 for example.
https://www.mersenne.org/report_expo...1085613&full=1
https://www.mersenne.org/report_expo...1079863&full=1
Or even worse, to (over 50) LL assignments, many without adequate bounds P-1 completed first. That is a problem, causing waste. Some by inadequate factoring; much more by LL + requiring a full eventual DC of some sort.
https://www.mersenne.ca/exponent/101780153
https://www.mersenne.ca/exponent/101308807
https://www.mersenne.org/report_expo...1805937&full=1
https://www.mersenne.org/report_expo...1835761&full=1
https://www.mersenne.org/report_expo...1857697&full=1
https://www.mersenne.org/report_expo...1886931&full=1
(and many more examples not listed)
Historically, some primality testing completes without any P-1 factoring ever being performed, and that is a waste of some time.
Anyone who does not understand why may read https://www.mersenne.org/various/math.php#p-1_factoring for an explanation.


edit: It looks like no more 101-102M P-1 assignments are available currently, based on a recent assignment batch for 1 gpu for 1 day yielding all 102.3M assignments.

[lisanderke]

Oops, it seems like I did what the OP of this thread didn't want to see, I used the manual GPU assignment tool to get Cat1 DC TF assignments... I grabbed about 70 exponents and I should be able to TF all of them within a matter of days (3, perhaps 4). I hope this is not an issue for anyone here! I'd be happy to unassign them if for whatever reason I'm not able to complete them within the time frame of a few days. As far as I understand it though, in this short amount of time I should not be hindering the DC effort as long as I turn in the results (relatively) quickly, no?


For future reference: would I be better off just getting manual GPU DC assignments just under 66M (current active range)? It seems a bit of a waste in my opinion if these TFs in the cat1 range aren't done and slower PCs are taking care of the LL DC. (A small look at the current active LL DCs shows a lot of them have not been TF'd to 75.)


Just to be clear: I do not intend to poach any active LL DCs, obviously.


Edit: Current TF DC range is max 64,682,541, whoops

[Runtime Error]

Originally Posted by kriesel

Some primality tests now in progress in 101M-102M have inadequate P-1 (stage 1 only, or low bounds), or none performed yet, yet have already progressed to PRP assignments.

What amount of B2 will Prime95 run with the default 0.3 GB of memory for second stage P-1?

[ATH]

Originally Posted by Runtime Error

What amount of B2 will Prime95 run with the default 0.3 GB of memory for second stage P-1?

@Runtime Error: Did you register 59 P-1 assignments on Jan 4th between 102,057,000 and 102,060,000 ? Until recently you still had those P-1 assignments with no progress on them, just like all those for Ryan Propper.

I am just curious if you registered those intentionally? I am checking to see if there is a bug or some other weird things going on.

[Runtime Error]

Originally Posted by ATH

@Runtime Error: Did you register 59 P-1 assignments on Jan 4th between 102,057,000 and 102,060,000 ? Until recently you still had those P-1 assignments with no progress on them, just like all those for Ryan Propper.

I am just curious if you registered those intentionally? I am checking to see if there is a bug or some other weird things going on.

Yes very likely, but anything from Jan 4th has been turned in already. My assignments page shows that my oldest assignment is from 2021-01-15. Those go onto "Manual Testing" machines, and I manually turn them in periodically.