mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data

Reply
 
Thread Tools
Old 2019-08-12, 08:00   #12
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

2·33·5·31 Posts
Default

Quote:
Originally Posted by kriesel View Post
Nobody is making exceptional claims about <exponent xxx>.
Just to note that some exponents in the past were used for different purposes (like benchmarks, etc) and they do have tens or hundreds of tests, for those reasons. For the rest, we are mainly in a violent agreement
LaurV is offline   Reply With Quote
Old 2019-08-12, 14:22   #13
ATH
Einyen
 
ATH's Avatar
 
Dec 2003
Denmark

2·13·107 Posts
Default

Quote:
Originally Posted by GP2 View Post
Note that TJAOI is currently finding factors of bit length 65.97 and should finish up to 66 very soon. Therefore there is not much point doing TF to less than 67 bits.
Yeah, he is the main reason I did not continue this project. Seems plausible he found all the factors below that limit.

It just annoys me those low bit levels are not showing up on the exponents...
ATH is offline   Reply With Quote
Old 2019-08-12, 15:03   #14
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

D5616 Posts
Default

Quote:
Originally Posted by LaurV View Post
Just to note that some exponents in the past were used for different purposes (like benchmarks, etc)
I suppose at one time during their development the available applications could have lacked timing or throughput display. That is no longer the case, for years now if I recall correctly.

The most commonly used GIMPS applications provide convenient measures of throughput or its inverse at the console or log file during production computations (with no need to duplicate previously completed work):
prime95/mprime ms/iter in LL or PRP, time taken between progress indications in P-1 stage 1 or 2 or in TF
Mlucas sec/iteration
gpuowl ms/it or ms/sq depending on version
CUDALucas ms/iter
Mfaktc GhzD/day
Mfakto GhzD/day
CUDAPm1 ms/iter or ms/transform
cllucas ms/iter (obsolescent, use faster gpuowl)

Benchmarking directions:
gpu LL test https://www.mersenne.ca/cudalucas.php upper right of page; 30,000 iterations of 57885161

gpu trial factoring https://www.mersenne.ca/mfaktc.php "show benchmark submission form" link near the top of that page; fill in the form with information gathered from a useful production run.

cpu LL & TF (prime95) benchmarks: https://www.mersenne.ca/bench.php
"Data shown here is mostly for historical reference, and is no longer being actively maintained."

There's no need now in benchmarking, for additional duplicative complete primality tests or P-1 runs or TF runs, nor to clutter up the primenet database with them. Timing information can be obtained from useful runs on exponents that need runs. There are also scaling guides available to estimate from one exponent's timing to a very different exponent's timing, either by the O(n log n log log n) or similar from number theory, or empirical relations developed by runs of widely spaced exponents that were not duplicated work.

Last fiddled with by kriesel on 2019-08-12 at 15:10
kriesel is offline   Reply With Quote
Old 2019-08-12, 15:32   #15
GP2
 
GP2's Avatar
 
Sep 2003

1010000100102 Posts
Default

Quote:
Originally Posted by ATH View Post
Yeah, he is the main reason I did not continue this project. Seems plausible he found all the factors below that limit.

It just annoys me those low bit levels are not showing up on the exponents...
We have only empirical evidence that TJAOI's search is exhaustive, but it's very strong empirical evidence. I posted about it a couple of years ago.

Perhaps Madpoo can change the Exponent Status report page on mersenne.org and James Heinrich can change the mersenne.ca page, to have a static fixed disclaimer that warns people against TFing less than 67 bits, especially for small exponents where the actual TF that was done is much lower than that.

And maybe the ECM status of each exponent could also be displayed on the Exponent Status page, with the t=N digits level translated into a bit level, again to warn about TF being fruitless.
GP2 is offline   Reply With Quote
Old 2019-08-12, 20:47   #16
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·3·569 Posts
Default

Quote:
Originally Posted by masser View Post
Some of the excessive duplication could be the result of people testing their hardware and/or software. The database is chock full of results that can be used to verify a user's setup.
The regular double-checks of LL and PRP first tests are useful cases of lengthy reliability runs. At one time the extra duplication may have been necessary for lack of a better self-test alternative, but copious duplication on small exponents are not likely to be as effective as the built-in self-tests now available in the most commonly used GIMPS applications. Duplicative self-tests should not be reported to the server if they are unnecessary and highly redundant relative to the prior effort of determining the exponent's nature.

The current status of the most common GIMPS applications is as I understand it,
  • prime95/mprime has a "torture test"
  • Mlucas -s (specify a size tiny small medium large huge or all)
  • CUDALucas -r (although this is limited to 8192k fft sizes maximum)
  • CUDAPm1 lacks the equivalent of -r, although it lists it in its help output. For this application the readme contains lists of various exponents' required bounds and known factors for testing software and hardware.
  • Mfaktc -st or -st2 or brief selftest on startup
  • Mfakto -st or -st2 or --cltest
  • GpuOwL PRP versions use the Gerbicz check. The very early LL version at one point had a separate test vector file so the test was reconfigurable.
  • cllucas -r (obsolescent LL program, use gpuowl, it's faster and reliable)
The builtin self-tests are likely to be faster in many cases, than duplicate exponent testing, and also won't clutter the mersenne.org or mersenne.ca servers' databases with extraneous duplicate records of the same test cases' many repetitions.

There are also collections of interim residues available online for comparison of short runs on a variety of exponent and therefore a variety of fft sizes. See https://www.mersenneforum.org/showpo...82&postcount=4 for LL interim residues, and https://www.mersenneforum.org/showpo...83&postcount=5 for PRP interim residues.
kriesel is offline   Reply With Quote
Old 2019-08-12, 21:15   #17
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

211648 Posts
Default

Quote:
Originally Posted by kriesel View Post
The current status of the most common GIMPS applications is as I understand it...
It is important to point out that the self-tests for mfaktX are for the software, not the hardware. And there have been cases where a card was bad and missed one or more factors (generally found by noticing a statistically significant low success rate in a large sample set)...
chalsall is online now   Reply With Quote
Old 2019-08-13, 01:45   #18
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

1101010101102 Posts
Default

Quote:
Originally Posted by chalsall View Post
It is important to point out that the self-tests for mfaktX are for the software, not the hardware. And there have been cases where a card was bad and missed one or more factors (generally found by noticing a statistically significant low success rate in a large sample set)...
Hmm, I'm not sure how you run or test the software without also somewhat exercising the hardware that the software runs on. Or did your particular sense of humor elude me again? But sure, you may want to also run a gpu memory tester or something to separate variables.
There are also cases where the software is ok and perhaps a gpu that's not will spit out the same bogus factor repeatedly, or when a Windows TDR hits on a very slow gpu there is a 1:1 correspondence; 12 TDRs, 12 occurrences of the same old factor of a different Mersenne number. Boosting the timeout value in the registry cleared it up, so good software, correctly functioning ancient gpu, just too slow for the default setting.

Last fiddled with by kriesel on 2019-08-13 at 01:53
kriesel is offline   Reply With Quote
Old 2019-08-13, 03:07   #19
srow7
 
Jul 2014

578 Posts
Default

@kriesel
maybe you could focus on more positive aspects of gimps
dont obsess about a 10 year old duplicate.
what i like to do is pick a subproject and contribute to it.
i dont know your skillset, here are some suggestions
1) beta test software
2) run benchmarks on new hardware
3) run cat0/1 LL/PRP to complete milestones
4) do gimps fundraising
5) high level gimps 5/10 year plan
6) try all work types (LL,PRP,TF,ECM,PM1) see which one you like
7) tune your hardware,software settings
8) install prime95 on an old laptop, see if it works

gimps is FUN!!!!
srow7 is offline   Reply With Quote
Old 2019-08-13, 18:57   #20
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

65268 Posts
Default

Quote:
Originally Posted by srow7 View Post
@kriesel
i dont know your skillset, here are some suggestions
1) beta test software
2) run benchmarks on new hardware
3) run cat0/1 LL/PRP to complete milestones
4) do gimps fundraising
5) high level gimps 5/10 year plan
6) try all work types (LL,PRP,TF,ECM,PM1) see which one you like
7) tune your hardware,software settings
8) install prime95 on an old laptop, see if it works
Thanks for taking the time to respond. Who is srow7, what are your favorite subprojects and contributions, & what is your skillset? I see you're currently #32 in top producers overall, which implies a certain commitment to the project. That list you gave above is a good start.

About me: see https://www.mersenneforum.org/showpo...3&postcount=11

See also https://www.mersenneforum.org/showth...922#post521922, begun May 2018, over 160 reference posts and growing. Have a look around there. Tactful suggestions for other additions are welcome.
kriesel is offline   Reply With Quote
Old 2019-08-13, 22:09   #21
srow7
 
Jul 2014

47 Posts
Default

I have tried all work types. I have 1 PC w/2 GPUs. I have a limited budget.
I like finding factors with mfaktc TF.
srow7 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Duplication of Effort for Smaller Aliquot Sequences EdH Aliquot Sequences 3 2018-04-17 13:31
Duplication of work: local vs. db EdH Aliquot Sequences 2 2010-12-31 04:30
2801^79-1; thoughts on duplication sampling fivemack Factoring 0 2010-04-15 22:23

All times are UTC. The time now is 21:20.

Sun Mar 29 21:20:41 UTC 2020 up 4 days, 18:53, 2 users, load averages: 1.52, 1.63, 1.58

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

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.