mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   News (https://www.mersenneforum.org/forumdisplay.php?f=151)
-   -   New Prime? (...NO) (https://www.mersenneforum.org/showthread.php?t=24102)

travisjank 2019-02-22 05:09

New Prime? (...NO)
 
Possible Prime?

M90,xxx,xxx might be a new prime! The admins have been notified and will investigate this result before it can be checked in.

Lucas-Lehmer test: !!!PRIME!!!

axn 2019-02-22 05:39

I have hidden the exponent. What makes you think you have found a prime? A "No factor" TF result doesn't mean the number is prime.

If you have completed a successful LL test or PRP test and got back a positive result, you can just check that result in to the server.

moebius 2019-02-22 08:17

He checked in a 83M ,a 89M and some double check LL-results in to the server today, maybe he is right.........

preda 2019-02-22 08:43

I can run a PRP with GpuOwl, expected time about 2days. Please PM me the exponent if desired.

penlu 2019-02-22 09:57

I can also offer verification via CUDALucas in 90 hours.

GP2 2019-02-22 11:56

Some [URL="https://www.mersenne.org/report_ll/?exp_lo=2&exp_hi=999999999&exp_date=&end_date=&user_only=1&user_id=travisjank&dispdate=1&B1="]recent results from TravisJank[/URL] have mismatches: [M]49769581[/M], [M]49867891[/M], [M]51237019[/M], [M]55863571[/M], and others have error codes marked Suspect: [M]88648837[/M], [M]88735033[/M], [M]88828589[/M], [M]88903769[/M].

The machine NG-TJ-P1 has two mismatches, two Suspect unverified results, and two verified good.

The machine NG-BTJ has one mismatch, one Suspect unverified result, and three verified good.

The machine DV-ROG has one mismatch, one Suspect unverified result, and one verified good.

Several other machines have clean track records.

Which machine found the possible prime?

kriesel 2019-02-22 15:31

confirmation time
 
If I recall correctly, confirmation run durations were of order 2 days to a bit under a day depending on hardware and software combination last time, for 82589933. Scaling to 90M, on same hardware and software, by p[SUP]2.12[/SUP], aspiring confirmation testers would need to have validly estimated run times of under about 2.4 days (~58 hours) to be competitive.
The fft file generated by CUDALucas 2.06 May 5 2017 build for an NVIDIA GTX 1080 Ti contains
(fftl(K) max exponent ms/iter)
5184 95507747 3.5816
so running a 90M LL (without Jacobi check; none implemented yet in CUDALucas) would take about 89.5 hours.

Best timing I have for gpuowl is V6.2, 4.206ms on an AMD RX480 at 5 M fft length, corresponding to a bit over 105. hours. (This would have a Gerbicz check so would be quite reliable.)

Dual-Xeon-e5-2670: 4.43 msec/iter in prime95 V29.4b8, so 110.75 hours.

I'd be willing to run verifications on any of these, but based on timings, it looks like I'm not in the running for confirmation of suspected primes.
I'd be happy to run a DC on a suspected false positive, on any of the above.
[url]https://www.mersenneforum.org/showpost.php?p=502243&postcount=29[/url]

Uncwilly 2019-02-22 15:51

George thinks the chances are this is a bad compile of a GPU program. He has gotten the save file from the OP and other info. Enjoy your weekend, nothing to see here.

Prime95 2019-02-22 16:06

Sorry, Travis. I just looked at the save file in a text editor and it appears to be all zeroes.

Please use the latest CudaLucas (which may be marked Beta) (2.06??), it is supposed to detect this problem. I have no idea why this important fix never made it into the production version.

ATH 2019-02-22 16:41

Yeah, you need to use CUDALucas "Beta" 2.06:

[url]https://sourceforge.net/projects/cudalucas/files/2.06Beta/[/url]

ET_ 2019-02-22 17:11

[QUOTE=ATH;509147]Yeah, you need to use CUDALucas "Beta" 2.06:

[url]https://sourceforge.net/projects/cudalucas/files/2.06Beta/[/url][/QUOTE]

Is there a 2.06 beta for Linux (or at least source code to compile)?


All times are UTC. The time now is 11:54.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.