New ECPP record (currently: 73,269 digits) - Page 7

frmky · 2022-12-14, 04:01

Quote:

Originally Posted by Batalov

If that process gets killed - you have to start from scratch because no data in the state file (.cert2) are written that are not in sequence from #0 to #tests. Maybe this can be refactored (thinking about it)?

That's no longer true in the current development version. That was a patch I sent to Andreas before embarking on these large runs.
https://gitlab.inria.fr/enge/cm/
https://gitlab.inria.fr/enge/cm/-/co...2b479edd45cd16

Batalov · 2022-12-14, 05:57

Quote:

Originally Posted by frmky

That's no longer true in the current development version.

Good!

paulunderwood · 2022-12-14, 06:25

Quote:

Originally Posted by Batalov

Good!

What would be "good" too is saving the class number calculations, even if it is gigabytes, because that can take several hundred core hours every time one starts stage 1. It would not take as long to read them into memory again on resumption.

frmky · 2022-12-14, 06:47

Quote:

Originally Posted by paulunderwood

What would be "good" too is saving the class number calculations, even if it is gigabytes, because that can take several hundred core hours every time one starts stage 1. It would not take as long to read them into memory again on resumption.

Also done.
https://gitlab.inria.fr/enge/cm/-/co...56d9c57e5e8454

Edit: For the run I just completed, the class numbers file is 32GB, and the primorials (also optionally saved and loaded) are 11GB.

Batalov · 2022-12-14, 06:51

Gigabytes are pennies these days. Our disks are necessarily over a petabyte (at a genome sequencing center).
Saving a temp image of a few gigabytes is certainly fine. Could draw a limit at, say, a terabyte or two.

Surely, the good idea is to start the ecpp.ini file with some limit settings. Everyone could set for themselves.

xilman · 2022-12-14, 10:41

Quote:

Originally Posted by frmky

125330^3+3^125330, at 59,798 digits, is prime.

Nice.

xilman · 2022-12-14, 10:49

Quote:

Originally Posted by Batalov

Gigabytes are pennies these days. Our disks are necessarily over a petabyte (at a genome sequencing center).
Saving a temp image of a few gigabytes is certainly fine. Could draw a limit at, say, a terabyte or two.

Surely, the good idea is to start the ecpp.ini file with some limit settings. Everyone could set for themselves.

Indeed.

Here is a GNFS Cado run, only half-way completed and using only two machines sieving. Chicken feed, IOW.

pcl@horus:~/nums/cado-nfs/work$ du -h
11M ./client/horus.work
11M ./client/horus+4.work
40M ./client/download
11M ./client/horus+5.work
9.7M ./client/horus+3.work
11M ./client/horus+2.work
11M ./client/horus+6.work
101M ./client
14G ./GW3_619.upload
4.7G ./GW3_619.dup1/0
4.7G ./GW3_619.dup1/1
9.3G ./GW3_619.dup1
26G .
pcl@horus:~/nums/cado-nfs/work$ df -h .
Filesystem Size Used Avail Use% Mounted on
/dev/sdc1 917G 235G 636G 27% /home
pcl@horus:~/nums/cado-nfs/work$

Essentially all of that is temporary files in that the relations are superfluous after the square root computation has finished.

paulunderwood · 2022-12-15, 19:50

Quote:

Originally Posted by frmky

Also done.
https://gitlab.inria.fr/enge/cm/-/co...56d9c57e5e8454

Edit: For the run I just completed, the class numbers file is 32GB, and the primorials (also optionally saved and loaded) are 11GB.

I found from the news file that you have to setup a directory and flag it to CM. Here is what I did in my ecpp directory

Code:

mkdir R_class
export CM_ECPP_TMPDIR="R_class"

(The estimation for the whole process to complete for R86453 is 1300 core hours. Reading back on resumption will take minutes.)

rudy235 · 2022-12-16, 01:45

The seven largest primes in the category ECPP have been discovered in the last 10 months (Mar-Dec 2022)

frmky · 2023-02-21, 18:46

I am aware this will soon be exceeded, but 104824^5+5^104824, at 73,269 digits, is prime. Stage 1 took 32 days on 20 24-core computers using GWNUM. Stage 2 took 27 days on 8 20-core computers. A few steps with large prime factors of h took most of the time in stage 2. I will explore the effects of further limiting the largest prime factor of h. Thanks again to Andreas for creating CM and Paul for adding support for GWNUM.

xilman · 2023-02-21, 21:27

Quote:

Originally Posted by frmky

I am aware this will soon be exceeded, but 104824^5+5^104824, at 73,269 digits, is prime. Stage 1 took 32 days on 20 24-core computers using GWNUM. Stage 2 took 27 days on 8 20-core computers. A few steps with large prime factors of h took most of the time in stage 2. I will explore the effects of further limiting the largest prime factor of h. Thanks again to Andreas for creating CM and Paul for adding support for GWNUM.

Nice work!

Any estimate when the first 100K digit prime will be proven?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
ECPP-DJ	danaj	Computer Science & Computational Number Theory	59	2020-10-10 04:57
Comparison to ECPP?	carpetpool	PARI/GP	2	2020-03-11 01:07
Can I just leave this here? (ECPP)	trhabib	Miscellaneous Math	6	2011-08-19 16:34
Looking for ECPP software	nuggetprime	Software	14	2010-03-07 17:09
new ECPP article	R. Gerbicz	GMP-ECM	2	2006-09-13 16:24

2022-12-14, 06:51	#71
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 10,093 Posts	Gigabytes are pennies these days. Our disks are necessarily over a petabyte (at a genome sequencing center). Saving a temp image of a few gigabytes is certainly fine. Could draw a limit at, say, a terabyte or two. Surely, the good idea is to start the *ecpp.ini* file with some limit settings. Everyone could set for themselves.

2022-12-16, 01:45	#75
rudy235 Jun 2015 Vallejo, CA/. 2203₈ Posts	The seven largest primes in the category ECPP have been discovered in the last 10 months (Mar-Dec 2022)

2023-02-21, 18:46	#76
frmky Jul 2003 So Cal A3D₁₆ Posts	I am aware this will soon be exceeded, but 104824^5+5^104824, at 73,269 digits, is prime. Stage 1 took 32 days on 20 24-core computers using GWNUM. Stage 2 took 27 days on 8 20-core computers. A few steps with large prime factors of h took most of the time in stage 2. I will explore the effects of further limiting the largest prime factor of h. Thanks again to Andreas for creating CM and Paul for adding support for GWNUM.