View Single Post
Old 2009-04-14, 16:49   #2
fivemack
(loop (#_fork))
 
fivemack's Avatar
 
Feb 2006
Cambridge, England

3·19·113 Posts
Default

In this particular case, all that's important is that you sieve the same range; you're getting roughly the same set of candidates out and just rejecting different proportions of them.

I get:

28-bit, 13: total yield: 1848, q=15001001 (0.04003 sec/rel)
28-bit, 14: total yield: 3966, q=15001001 (0.04397 sec/rel)
27-bit, 13: total yield: 922, q=15001001 (0.07872 sec/rel)
27-bit, 14: total yield: 1965, q=15001001 (0.08762 sec/rel)

and with small prime bounds at 12 million (note I'm still searching Q around 15M)

28-bit, 13: total yield: 1706, q=15001001 (0.03774 sec/rel)
28-bit, 14: total yield: 3646, q=15001001 (0.04481 sec/rel)
27-bit, 13: total yield: 856, q=15001001 (0.07520 sec/rel)
27-bit, 14: total yield: 1824, q=15001001 (0.08388 sec/rel)

and just try one with small prime bounds at 9 million since 12 was better than 15

28-bit, 13: total yield: 1505, q=15001001 (0.03880 sec/rel)

So: siever 13e seems to be the right one, small prime bound 12M looks good, and since ~2 relations per Q appears to be a sweet spot, I'd use 28-bit large primes. The job will take under a million CPU-seconds, say one quad-core for four days.
fivemack is offline   Reply With Quote