Using 16e on smaller numbers

fivemack · 2017-09-19, 07:28

I've just finished sieving the log=185.44 122!+1 cofactor with 16e/32LP. 350M relations (295M unique) from sieving 30M-132.4M (because I wanted a multiple of 128 jobs given my cluster) gave a 15.77M matrix at density 124; this is much easier than the 26.1M density-120 matrix that nfs@home got with 365M/291M relations at 15e/32 on the only slightly harder 114!+1 cofactor.

A bit slower sieving per relation

Code:

wheat@wheat:/home/nfsworld/factorial/F122p/tt$ cat 16e-22.t
total yield: 35827, q=134010001 (0.69377 sec/rel)
wheat@wheat:/home/nfsworld/factorial/F122p/tt$ cat 15e-22.t
total yield: 16056, q=134009999 (0.61100 sec/rel)

but I think still a win

pinhodecarlos · 2017-09-19, 07:45

And you have more cores connected to 16e V5 since it gives more points so it will be sieved even faster.

What about memory allocated by thread? People are complaining about the higher usage of memory when you run the 15e application with the 16e one.

Edit: oops, though this thread was on the NFS@Home.

VBCurtis · 2017-09-19, 08:29

Is it possible the new matrix was easier in part because you chose lower alim/rlim for your 16e run than NFS@home chose for 15e?

I've been wondering how (if?) matrix size scales with alim/rlim choice, holding LP and input difficulty fixed. Specifically, wondering if some of the jobs presently going to 14e with alim = 268M would produce smaller matrices on 15e with alim = 134M or thereabouts.

fivemack · 2017-09-19, 08:52

Quote:

Originally Posted by VBCurtis

Is it possible the new matrix was easier in part because you chose lower alim/rlim for your 16e run than NFS@home chose for 15e?

I've been wondering how (if?) matrix size scales with alim/rlim choice, holding LP and input difficulty fixed. Specifically, wondering if some of the jobs presently going to 14e with alim = 268M would produce smaller matrices on 15e with alim = 134M or thereabouts.

That's certainly conceivable (the 16e run had alim=134M). I would say that most of the larger jobs on 14e at the moment ought to be on 15e, but having three levels of difficulty is apparently useful for attracting BOINCers.

2017-09-19, 07:28	#1
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 1100011101011₂ Posts	Using 16e on smaller numbers I've just finished sieving the log=185.44 122!+1 cofactor with 16e/32LP. 350M relations (295M unique) from sieving 30M-132.4M (because I wanted a multiple of 128 jobs given my cluster) gave a 15.77M matrix at density 124; this is much easier than the 26.1M density-120 matrix that nfs@home got with 365M/291M relations at 15e/32 on the only slightly harder 114!+1 cofactor. A bit slower sieving per relation Code: wheat@wheat:/home/nfsworld/factorial/F122p/tt$ cat 16e-22.t total yield: 35827, q=134010001 (0.69377 sec/rel) wheat@wheat:/home/nfsworld/factorial/F122p/tt$ cat 15e-22.t total yield: 16056, q=134009999 (0.61100 sec/rel) but I think still a win Last fiddled with by fivemack on 2017-09-19 at 07:29

2017-09-19, 07:45	#2
pinhodecarlos "Carlos Pinho" Oct 2011 Milton Keynes, UK 11370₈ Posts	And you have more cores connected to 16e V5 since it gives more points so it will be sieved even faster. What about memory allocated by thread? People are complaining about the higher usage of memory when you run the 15e application with the 16e one. Edit: oops, though this thread was on the NFS@Home. Last fiddled with by pinhodecarlos on 2017-09-19 at 07:49

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
NFS on smaller numbers?	skan	YAFU	6	2013-02-26 13:57
Sequences with smaller cofactors	Mr. Odd	Aliquot Sequences	8	2010-12-01 17:12
Smaller filtering run oddity	10metreh	Msieve	17	2009-01-05 14:58
checking smaller number	fortega	Data	2	2005-06-16 22:48
Factoring Smaller Numbers	marc	Factoring	6	2004-10-09 14:17

2017-09-19, 08:29	#3
VBCurtis "Curtis" Feb 2005 Riverside, CA 1215₁₆ Posts	Is it possible the new matrix was easier in part because you chose lower alim/rlim for your 16e run than NFS@home chose for 15e? I've been wondering how (if?) matrix size scales with alim/rlim choice, holding LP and input difficulty fixed. Specifically, wondering if some of the jobs presently going to 14e with alim = 268M would produce smaller matrices on 15e with alim = 134M or thereabouts.