Ten Billion Digits Mersenne Numbers

[James Heinrich]

Originally Posted by LaurV

I have files with ALL factors for Mp with p<10 billions (ten times the range of GIMPS, includes the 3.x billions of the OBD range) for up to 49 bits or so, many megabytes and few months of work.

If you're willing to share that with me, I'd be very interested, especially in the M(10⁹)-M(2³²) range, but if you want to send me all I'm sure I'd find them useful later.

My PHP programming knowledge is reasonably good, but my math knowledge is very very poor, so I'm no good at PARI programming because, well, I don't know what I'm doing

[stathmk]

I must be missing some information. What programs besides GIMPS are you using? I want to find out if there’s a program that can find at least one factor of 2^((10^19-1)/9)-1 and/or 2^((10^23-1)/9)-1. I wondered if I should check a C or Cygwin program before going onto GIMPS. I’ve been checking Operation Billion Digits at web pages like these:
http://www.mersenneforum.org/showthread.php?t=15294
http://home.earthlink.net/~elevensmo....html#Download
http://www.moregimps.it/mersenne-test/howto.php
http://www.moregimps.it/billion/download1.php (I can’t get Cygwin to work for some reason. I know I got it working 4 years ago.)
About how many weeks would it take to find a factor?
What do you suggest that I do?

Edit: At Cygwin setup, I'm not understanding which packages to download.

[aketilander]

Originally Posted by stathmk

I must be missing some information. What programs besides GIMPS are you using?

http://www.moregimps.it/billion/download.php

[LaurV]

Sorry for necro-posting, somebody found my code above and liked a part of it, probably because "it looks complicate", haha, and sent me a PM, asking for the newest version. I already forgot about these "experiments" or how should I call them, and sometimes when new people join the forum and walk the same path, nostalgia catches me. The code is ancient, and in the stage (bitlevels) the TF is now, it is totally futile for both Gimps (<1G) and James' (<10G) databases.

Even at the time, the code was posted for didactic purpose, if you want to play with pari/gp or learn some math, check how the factors look and how they can be found, etc. Below 32 bits expos (4.29G), mfakt[c|o] is about a million times faster, and between 4G29 and 10G, one can use factor5, or other pieces of code.

If you still want to play with pari/gp TF, I posted few pieces of code in another thread (post 32, 35, 52, but you should read all the discussion) which split the factor candidates in classes (similar to mfakt[c|o]), which is also much faster, and which code was also improved by another user (hansl) who wrote a "paralelized" version to use all cores (post #118 and the discussion that follows).

For pari/gp P-1, there was also a series of posts in the past, with much faster code, (I can't find them**, but the only one I know of, which is able to properly extend B1 limit, assuming you kept the saved file, is attached to this post from 2013). That code always uses E=1 and d=6, as Carl (owftheevil) comments further, it doesn't do any Brent-Suyama extension, but it uses almost no memory for stage 2, and it still can find quite large factors, but compared with state of the art actual programs (CPU or GPU), it is way alower, and also futile, beside of didactic scope.

My "admirer" asks also what becomes of the cuda version. Well...nothing. There is a cuda version for it, but it only does stupid multiplication, so it is slow and it can only be used for small exponents, which is also futile. My skill in implementing FFT multiplication (cuda or not) is almost non-existent (albeit I understand the math reasonably well).

-------------
** It took me about 20 minutes to find all the links I used in this post, I said so many times I will collect some of the best posts (mine and others') in my blog, like kriesel is doing, to be easy to refer later, but being to busy or too lazy impeded me till now. In fact, I can see this thread is so old, that many of the links posted here by other users are already dead too, for example the link Ake refers above is to Luigi Morelli's double mersenne and/or fermat numbers (moreware and moregims stand for Morelli, ), he got proper domains for them and there is no more-whatever any-more .