mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Marin's Mersenne-aries (https://www.mersenneforum.org/forumdisplay.php?f=30)
-   -   processed dc and tc posts (https://www.mersenneforum.org/showthread.php?t=24152)

kriesel 2019-08-16 19:59

[QUOTE=Uncwilly;523767]Can you PM GP2 about this and work it out? My current knowledge about residue types, etc. is nearly non-existent.[/QUOTE]
Will do. I see you did the post one edit I suggested, and more. One bit of lint:
sentence five there, "show [B]nor[/B] progress" I think you mean [B]no[/B] not [B]nor[/B].

Uncwilly 2019-08-30 04:04

[QUOTE=Uncwilly;524799]Also, normally people don't post about finishing, unless the OP of the exponent makes mention that they are concerned about a machine's stability..[/QUOTE][QUOTE=mrh;524816]Ok. I tried in a post on 8/15. Oh well, I'll go try something else.[/QUOTE][QUOTE=nomead;524826]Yeah but in addition to posting, you also need to register the assignment in Primenet.[/QUOTE]To make things clearer: Yes, you stated that you would run the number. But, since it wasn't registered on PrimeNet, progress on it was not able to be tracked. When you posted on 8/15 that it was completed, I did not register it as being significant in my mind, because nominally we don't report completion (as noted above.) I had moved the other post with the claim over to the other thread. There was no intentional slight.

Registering exponents prevents duplication of effort.

nomead, it registering is important, but checking in the results is sufficient to clear it.

mrh, Hang around and do what you find fun.

moebius 2019-10-27 16:55

1 Attachment(s)
A leftover last LL exponent of my flaky Phenom processor should be verified sooner. (Error ratio see PDF document). Maybe somebody wants to take this one.

[URL="https://www.mersenne.org/report_exponent/?exp_lo=81269471&exp_hi=&full=1"]81269471[/URL]

ewmayer 2019-12-27 03:45

1 Attachment(s)
[url=https://www.mersenne.org/report_exponent/?exp_lo=96365419&full=1]96365419[/url] is my first non-DC PRP test using Mlucas v19 ... this ran on my unstable Haswell, lots of random-error quits, but now I always run two 4-thread jobs at same time on that machine, if one dies other can continue. Logfile attached, the timings ~17ms/iter were with job getting ~90% cycles, the ones ~30 ms the job was getting ~50% cycles on this system. Early in the run you can see it emitting ROE = 0.4375 warnings and getting stuck in an infinite loop, that was the reason for my 3 Dec patch to the original 1 Dec v19 releae. So lotsa logfile mess, but no Gerbicz-check errors. An early DC using a different program would be appreciated.

kriesel 2019-12-27 13:25

[QUOTE=ewmayer;533615][URL="https://www.mersenne.org/report_exponent/?exp_lo=96365419&full=1"]96365419[/URL] is my first non-DC PRP test using Mlucas v19 ... this ran on my unstable Haswell, lots of random-error quits ... An early DC using a different program would be appreciated.[/QUOTE]Which PRP residue type does Mlucas do? Never mind, I see it was type 1. But can't reserve that exponent for DC. Looks like endless_mike is on it.

endless mike 2020-01-07 02:04

[QUOTE=ewmayer][M]96365419[/M] is my first non-DC PRP test using Mlucas v19 ... this ran on my unstable Haswell, lots of random-error quits, but now I always run two 4-thread jobs at same time on that machine, if one dies other can continue. Logfile attached, the timings ~17ms/iter were with job getting ~90% cycles, the ones ~30 ms the job was getting ~50% cycles on this system. Early in the run you can see it emitting ROE = 0.4375 warnings and getting stuck in an infinite loop, that was the reason for my 3 Dec patch to the original 1 Dec v19 releae. So lotsa logfile mess, but no Gerbicz-check errors. An early DC using a different program would be appreciated.[/QUOTE]

[QUOTE=endless mike]I just added it and was coming back to state my claim.[/QUOTE]

[QUOTE=ewmayer]Thanks - I woke up during the night and realized I'd forgot to mask off the last 2 digits of the final Res64 in the logfile I posted, so strictly as a matter of form - this is a scientific project, after all - I suggest we impose on George by having you send him your final checkpoint file when your run completes, so he can re-run the last few thousand iterations and validate the result, like he does for new-prime cases reported to the server.

(That is assuming your result matches mine - if not, it won't be any issue, and instead we'll be needing a 3rd run).[/QUOTE]

It looks like we matched. If someone could explain to me what to send to George, that would be great.

ATH 2020-01-07 04:56

[QUOTE=endless mike;534455]It looks like we matched. If someone could explain to me what to send to George, that would be great.[/QUOTE]

The newest p96365419.* file, so the last iterations can be rerun. The problem is the file is around 45-48 MB so cannot be sent by email, not sure if you have a way to send it or host it.

endless mike 2020-01-19 02:13

Claimed this one.
Is there a post that describes what all the fields are in a worktodo entry for PRP? I always guess and hope I did it right.

[QUOTE=GP2;511979]PRP checks

EWM: My Haswell-quad run of [url=https://www.mersenne.org/report_exponent/?exp_lo=96365519&full=1]96365519[/url] was hit by the Mlucas v19 PRP-residue postprocessing bug reported [url=https://mersenneforum.org/showthread.php?t=25117]here[/url]. I've re-run the final Kiter using a new build with the bugfix and am 99.9999999% sure that the actual PRP-residue data were unaffected, but better safe than sorry ... and we can use more PRP-DC data for error-stats-gathering anyway.[/QUOTE]

ewmayer 2020-01-19 02:57

[QUOTE=endless mike;535486]Claimed this one.
Is there a post that describes what all the fields are in a worktodo entry for PRP? I always guess and hope I did it right.[/QUOTE]

Thanks - here is my original workdtodo entry, I presume/hope you can simply re-use it, i.e. assignment ID doesn't change from 1st-time test to DC:
[i]
PRP=EDDC25414116177C4F046D79BE11A463,1,2,96365519,-1,76,0
[/i]
Also, when you say you claimed it, did you actually reserve it via the server, or just by way of raise-hand-here, since the server won't be handing any such large ones out as DCs any time soon?

ewmayer 2020-01-19 19:25

[QUOTE=ATH;535506]PRP=EDDC25414116177C4F046D79BE11A463,1,2,96365519,-1,76,0,3,1

Added the 2 extra arguments that can be in the assignment: ",3,1"[/QUOTE]

Ah, thanks for reminding me of that difference between PRP and PRP-DC, Andreas.

Oh, re. reserving specific exponents for early-DC or (early whatever), the following occurred to me last night, wanted to run it by the folks here to gauge if it would be useful - on a given exponent's status page, if exponent has not been retired (via successful DC), enable a "reserve this exponent for [insert whatever work makes most sense for it]" widget on the status page.

(Actually, with deeper p-1 always being a possibility for not-completely-factored numbers and PRP-C now also being possible for same, it seems the only way to truly 'retire' a non-prime M-number or to probable-completely-factor it, i.e. 1 or more small factors found and PRP-C finds the remaining cofactor to be probable-prime.)

ewmayer 2020-03-05 20:00

[QUOTE=Uncwilly;538926]You just edited a post (no post saying there was fresh work) and did not put the work in a code box. Thus, no one noticed.[/QUOTE]
Ah - I figured the regulars around here would regularly scan the to-do entries in the various placeholder posts and grab the ones that fit their preferences. In future shall supplement basic assignment-needs-DC with new posts containing my long accompanying rambles. :) (Yah, I see your editorial snark).

[QUOTE]I have been off on holiday and then laid low by "what's going around". I had planned to do maintenance work on the lists.[/QUOTE]
Given the Covid-19 pandemic, these days one needs to be careful with verbiage like "I cought what's going around," lest one unduly alarm one's friends and loved ones.

Thanks to Jan S for taking up the 2 96M PRP-3 early-DCs.


All times are UTC. The time now is 02:20.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.