mersenneforum.org How Much Memory at Various Sizes?
 Register FAQ Search Today's Posts Mark Forums Read

 2005-04-19, 15:16 #1 wblipp     "William" May 2003 Near Grandkid 3·7·113 Posts How Much Memory at Various Sizes? How much memory does GMP-ECM use for default stage 2 for ECM at sizes from 50 digits to 70 digits? How about P-1 and P+1? Is there a simple way to find this out?
 2005-04-23, 18:47 #2 akruppa     "Nancy" Aug 2002 Alexandria 2,467 Posts Paul Zimmermann wrote an estimate for the memory use which is in the CVS version. It looks like this: Code:  lgk = ceil_log2 (dF); mem = 9.0 + (double) lgk; #if (MULT == KS) mem += 24.0; /* estimated memory for kronecker_schonhage */ mem += 1.0; /* for the wrap-case in PrerevertDivision */ #endif mem *= (double) dF; mem *= (double) mpz_size (modulus->orig_modulus); mem *= (double) mp_bits_per_limb / 8.0; To estimate dF: B2 - B2min = k * dF * d and d/dF ~= 11...12 for large B2. Lets assume 12. So dF ~= sqrt((B2-B2min)/k/12). I'm a little surprised that the memory requirements for Kronecker-Schönhage are that high... Alex
 2005-04-23, 23:21 #3 wblipp     "William" May 2003 Near Grandkid 3·7·113 Posts Hmmm. Could I get a worked example to help me understand how to apply this? I'm considering plans to work on the 244 digit composite 5349-1 and the 246 composite 19193-1, possibly at B1=260M or 850M. In preparation for this, I'm considering upgrading my PC to have 1, 2, or 3 Gigabytes of RAM. How much is "enough"? William
 2005-04-24, 09:03 #4 dave_dm   May 2004 24×5 Posts Something I've been thinking about: The -treefile option stores the product tree of F on disk; this reduces the memory consumption from O(dF * log(dF) * log(n)) to O(dF * log(n)) at the cost of more expensive IO. However with a threaded approach we could run a new stage 1 at low priority during a (-treefile) stage 2; this should 'use up' most of the wasted cycles doing disk seeks. So IMO, ecm probably doesn't require lots of memory, it's just easiest to implement that way. Dave
 2005-04-24, 17:42 #5 akruppa     "Nancy" Aug 2002 Alexandria 2,467 Posts Since B1=B2min<
 2005-04-24, 20:04 #6 Mystwalker     Jul 2004 Potsdam, Germany 3×277 Posts Xyzzy did some test for B1=26e7 here. I've also experienced that the treefile option is not really that bad for performance. I'd guess it costs less speed than a higher k value - I did no in-depth testing, though.

 Similar Threads Thread Thread Starter Forum Replies Last Post ixfd64 Hardware 4 2011-12-14 21:24 TimSorbet Software 7 2008-01-14 17:33 Cruelty Riesel Prime Search 3 2006-07-12 15:15 Dresdenboy Math 2 2003-12-29 22:55 Unregistered Hardware 4 2003-12-06 13:43

All times are UTC. The time now is 02:56.

Mon Jan 30 02:56:50 UTC 2023 up 165 days, 25 mins, 0 users, load averages: 1.53, 1.21, 1.08