Let's finish DC of old "dubious" claims of primes

[tServo]

My last 4 are in the books.
I finished them much earlier than I originally thought because I thought I queued some work before them.

[kriesel]

Originally Posted by tServo

My last 4 are in the books.

Thanks for your efforts.

Originally Posted by kriesel

Per https://mersenneforum.org/showpost.p...2&postcount=20 tServo will do these four, probably completing by mid January:
PRP=1,2,97970567,-1,76,0 Kriesel did first PRP shifted type 1
PRP=1,2,103404713,-1,76,0 Kriesel did first PRP shifted type 1
PRP=1,2,108575167,-1,77,0 Kriesel did first PRP shifted type 1
PRP=1,2,110000017,-1,78,0 Kriesel did first LL shifted

Server shows no indication of reservation or result report for those yet. Please report your results and upload the proof files.

[tServo]

Whoops. I worked on the wrong set.
I will redo if you want or if you don't, I'll understand.

[kriesel]

Try again.

[kriesel]

Per https://mersenneforum.org/showpost.p...2&postcount=20 tServo will do these, probably completing in January:
PRP=1,2,103404713,-1,76,0 Kriesel did first PRP shifted type 1
PRP=1,2,108575167,-1,77,0 Kriesel did first PRP shifted type 1

These are available:
PRP=1,2,231695603,-1,79,0 Kriesel did first PRP shifted type 1
PRP=1,2,276036017,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,285754177,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,315690521,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,369452123,-1,81,0 Kriesel did first PRP unshifted type 1
PRP=1,2,852348659,-1,86,0 Kriesel did first PRP unshifted type 1
(those exponents not present in this update have a completed PRP/proof & Cert)
Only 6 more to claim for PRP with proof-generation, each of which will take a while.
(Use prime95/mprime v30.3 or higher, or Gpuowl 6.11-318 or higher for PRP/proofgen.)

M333333223 Kriesel PRPDC/GEC/proof-gen completed, and a cert will be needed. I think this will eventually appear in the special Cert assignments portion of https://mersenneforum.org/showpost.p...79&postcount=4.

[tServo]

I should be finished.

[kriesel]

Only 6 more to claim for PRP with proof-generation, each of which will take a while. These are available:
PRP=1,2,231695603,-1,79,0 Kriesel did first PRP shifted type 1
PRP=1,2,276036017,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,285754177,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,315690521,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,369452123,-1,81,0 Kriesel did first PRP unshifted type 1
PRP=1,2,852348659,-1,86,0 Kriesel did first PRP unshifted type 1
(Use prime95/mprime v30.3 or higher, or Gpuowl 6.11-318 or higher for PRP/proof-gen. Be sure to upload the proof files if running gpuowl manually.)

Originally Posted by tServo

I should be finished.

Thanks for the help. Please upload the proof files for
PRP=1,2,103404713,-1,76,0
PRP=1,2,108575167,-1,77,0
so that the CERT runs can be issued as assignments. (It was discussed and decided when PRP with proof was being added as a GIMPS capability, that even with matching PRP 64-bit residues, if a proof file is available for the second PRP, the Cert will be run.)

(those exponents not present in this update have a completed PRP/proof & Cert)

[kriesel]

Originally Posted by tuckerkao

The only oversized exponent seems to be M852348659

Not exactly, if Mp predictions are considered. The 5 largest exponent Mp predictions on record without DC or PRP & successful CERT completed are:

• Mp#52* prediction list M402143717 I have a PRP/proof-gen in progress in a manual gpuowl run
• Mp#45 prediction list M655685803 (requires George's manual intervention due to the mersenne.org server's ~595M proof handling limit)
• appearing first as a dubious claim, then an Mp#52* prediction entry M852348659 (requires George's manual intervention due to mersenne.org server's ~595M proof handling limit)
• Mp#52* prediction list M1000000007 (1G+7) all advisable TF and P-1 factoring have been completed; the exponent is just beyond mersenne.org's 1G maximum, but within prime95's AVX512 PRP & CERT capabilities on a fast AVX512 system; the initial PRP with proof gen run would require ~5 months run time on a Radeon VII with reasonably recent/efficient gpuowl, and special manual handling to perform the CERT;
• Mp#51* prediction list M2252945257 (~2.253G); I have its TF in progress with months to go on a GTX 1080 Ti; P-1 is feasible in Mlucas V21.1 but premature at this time and so unstarted; to my knowledge there currently exists no software suitable, for PRP with proof generation, or for CERT handling, on such a large exponent. PRP will not be attempted without those capabilities. LL will not be attempted, unless there is first an extremely unlikely (~25. ppb chance) PRP-P result and corresponding successful CERT somehow. (It may be Mlucas, V22? that first provides the PRP/proof-gen capability for this exponent. For CERT there are no apparent near-term prospects. CERT would require manual upload/download skipping the mersenne.org server, and extension of prime95's max fft length to about twice its current AVX512 maximum, or perhaps a Karatsuba kludge atop the 64M fft length. That seems unlikely and not a good use of George's time for such a rare use case.)

The suggestion of capping Predict Mp#x* exponent predictions to no more than n times the exponent of Mp#(x-1)* where n is ~4, and also, to no more than 10⁹, sounds right to me for the next few x. Or perhaps cap exponent predictions at ~595M for x up to ~55, to make less work for George.

[kriesel]

I think these Mlucas versions are a little closer to the mark.

Originally Posted by kriesel

Mp#51* prediction list M2252945257 (~2.253G); I have its TF in progress with months to go on a GTX 1080 Ti; P-1 is feasible in Mlucas V21.1 V20.1.1 but premature at this time and so unstarted; ... (It may be Mlucas, V22? V21.x that first provides the PRP/proof-gen capability for this exponent. For CERT there are no apparent near-term prospects.

[JCoveiro]

Hi Kriesel.

Can I help with double-checks?

[kriesel]

Originally Posted by kriesel

Only 6 more to claim for PRP with proof-generation, each of which will take a while. These are available:
PRP=1,2,231695603,-1,79,0 Kriesel did first PRP shifted type 1, repeating with proof generation
PRP=1,2,276036017,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,285754177,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,315690521,-1,80,0 Kriesel did first PRP shifted type 1
PRP=1,2,369452123,-1,81,0 Kriesel did first PRP unshifted type 1
PRP=1,2,852348659,-1,86,0 Kriesel did first PRP unshifted type 1
(Use prime95/mprime v30.3 or higher, or Gpuowl 6.11-318 or higher for PRP/proof-gen. Be sure to upload the proof files if running gpuowl manually.)
(those exponents not present in this update have a completed PRP/proof & Cert or PRP DC)

Originally Posted by JCoveiro

Can I help with double-checks?

Sure. The list is open to anyone with the needed hardware and patience. I am currently running 231695603 again with proof generation on a fast GPU, ETA a few more days. (Could not reserve it on mersenne.org since I did the first run.)

Feel free to reserve 276036017 or whatever other suits you, at https://www.mersenne.org/manual_assignment/, and run it on a reasonably efficient version of Gpuowl, with proof power ~9 on your GPU, or on a powerful CPU in mprime/prime95 V30.4 or newer, and plenty of available disk space for the temporary files, and post here that whichever exponent you choose has been claimed. A rough estimate of run time is ~2 weeks nonstop 24/7 for ~332M exponent PRP/GEC/proof on Radeon VII, ~20% longer on RX 6900XT, time varying as p^2.1. When complete, report the results and upload the resulting proof file to the PrimeNet server. Easiest for gpuowl submission is use of Preda's primenet.py, but manual results page for the result line and drag and drop the proof file into a running primenet-connected mprime or prime95's folder works too. Gpuowl v6.11-357 to -380 or v7.2-53 or v7.2-86 would be good choices. Maybe benchmark which version is fastest for your setup in the necessary fft length.
A rough estimate of 276036017 PRP duration is (276036017 / 332200000)^2.1 x 15 days ~ 10 days on Radeon VII, ~12 days on RX 6900XT.
At the other extreme 852M, ~108. days on Radeon VII, ~130. days on RX 6900XT.
Even the smallest of these could take months or years on a medium speed CPU. 852M would be ~3.5 YEARS on an i5-1035G1; 276M would be ~4 months on that.