Prime95 30.8 (big P-1 changes, see post #551)

[James Heinrich]

Originally Posted by alpertron

I tried to upload it manually, but the server responded the following

Thanks, fixed.

[Stef42]

Originally Posted by petrw1

Seems to have stopped occurring.
I cleaned up the disk; it was under 100K free.

Had this one last week, with 1 hour interval. Than a BSOD. Ran Memtest and did extensive testing, getting errors right away. Updated the BIOS (AM5 platform) and all errors gone. No issues in p95 since.
Not implying anything, but might worthwhile running Memtest for 24h?!

[kriesel]

Had a disk-full error on a system a while back, and lots of GIMPS apps were choking on the multi-gpu system. Mlucas on WSL skipped output of a progress line in its OBD P-1 stat file, and continued onward otherwise unphased, while I cleaned out old files to free up enough space for other apps to be restarted usefully.

[James Heinrich]

Originally Posted by alpertron

I ran PRP of M281903623 / known 3 prime factors using Prime95 version 30.8 build 17

Dario got the proof uploaded, but I've been keeping an eye on it and the proof still hasn't been certified. The cert has been assigned at least 4 times already, all to different anonymous users, none of them have completed it. Should there perhaps be some mechanism in place that if a cert has failed to complete more than twice it should be preferentially assigned to a "high-reliability" user rather than any random anonymous user?

[Prime95]

Big CERTs are sporadically made available here: https://www.mersenneforum.org/showthread.php?t=24148

Only dedicated users will get the assignment.

[kriesel]

Originally Posted by Prime95

sporadically

Yes indeed.
And exponents above ~595.8M exponent (the reach of prime95/gwunum SSE2's maximum 32M fft) require additional special handling by George on at least an AVX2 (up to ~920.8M exponent) or sometimes AVX512 system (above ~920.8M exponent).
Large exponent prp tests that previously had long waits for proof processing to generate cert assignments and successful cert completion include:
623466917 ~2.5 months
654000059 ~2.5 months
800011213 ~2 months
819751061 ~6 weeks
843112609 ~5 months
These are short compared to the multi-year waits common for LLDC, but long compared to the typically hours turnaround on PRP proof certs at the wavefronts.

Currently pending large-exponent certs (proof submitted as below, cert assignment not yet available for reservation) include
652589933 since 2023-04-26
937156667 since 2023-04-06; it's unclear to me how long completed proof submissions remain valid after upload, currently, before expiration due to a lack of cert. Certification can not be performed before a cert assignment is generated and available for the given exponent and proof file. (Per https://mersenneforum.org/showpost.p...postcount=2254 the proof file upload must occur within 60 days of the result submission. In this case it occurred same day.)

Refer also to https://www.mersenneforum.org/showpo...5&postcount=12

[Andrew Usher]

Someone took alpertron's CERT (am I correct that those assignments to 'Anonymous' were bogus and never really existed?) and now completed it, showing the proof good.

Kriesel, those exponents are hardly urgent, and it is arguably a gesture of goodwill that George does them at all. You should stop complaining about it. Certainly (and I am no more intending to divert the thread than you apparently were) it's less important that adding the Fermat numbers to the database (https://www.mersenneforum.org/showpo...9&postcount=36) which would also be less work, as the entries are already there.

[James Heinrich]

Originally Posted by Andrew Usher

Someone took alpertron's CERT (am I correct that those assignments to 'Anonymous' were bogus and never really existed?) and now completed it, showing the proof good.

Not bogus, manually reserved for those who choose to grab them. In this case:

Originally Posted by Naegi Makoto

Took
CERT=00000000000000000000000010CD8207,1,2,281903623,-1,1101187

[Mark Rose]

Is it possible to configure mprime to only run stage 1 of P-1?

I see OnlyRunStage2WithMaxMemory=1, which will kind of work, but is there a better way?

I'm trying to finish up some P-1 work before the end of the month, but only one of two machines I still have going has 128 GB of memory.

[storm5510]

Originally Posted by Mark Rose

Is it possible to configure mprime to only run stage 1 of P-1?

I see OnlyRunStage2WithMaxMemory=1, which will kind of work, but is there a better way?

I'm trying to finish up some P-1 work before the end of the month, but only one of two machines I still have going has 128 GB of memory.

With Prime95 when running ECM's, matching the B1 and B2 boundaries worked for me. This might work for P-1. You would just have to try it.

[Mark Rose]

Originally Posted by storm5510

With Prime95 when running ECM's, matching the B1 and B2 boundaries worked for me. This might work for P-1. You would just have to try it.

These are 332M exponents. I'm forcing a large B2 on a "low" memory machine to get similar work done as on the 128 GB machine, which is progressing very slowly. Yes, I know that's not optimal.

I've put in some configuration that should hopefully have the low memory machine skip Stage 2 so I can move those assignments to the other machine. The high memory machine will be in storage a little before it can finish the work I'd like to finish.

What work? Running P-1 on all the 332M+ exponents that have an LL or PRP result without having a P-1 result. Is it needed? Nope.