20100910, 23:32  #34 
"Bob Silverman"
Nov 2003
North of Boston
1D48_{16} Posts 
I will try both a quicksort and a hash table......

20100913, 16:48  #35 
Oct 2006
vomit_frame_pointer
360_{10} Posts 
Duplication data: 3LP vs. 2LP
I should have computed this a couple of weeks ago  moving will do that to a guy.
This was for 6^353+1 C178 GNFS. See earlier in the thread for parameters and filtering log. I included on the relations 52M < Q < 164M, as these were the ones sieved with the 3LP approach. Among the 2LP subset, the duplication rate was 14.6%: Code:
Found 102240374 unique, 17440243 duplicate, and 0 bad relations. Code:
Found 150125345 unique, 34802799 duplicate, and 0 bad relations. I'm running a C179 (5^418+1) as a 3LP job. I will have the duplication rates for the whole set vs. those for the 2LP subset this time. Should be available just before November 1st. 
20100920, 22:11  #36  
"Bob Silverman"
Nov 2003
North of Boston
1110101001000_{2} Posts 
Quote:
that I use inside NFS. Surprisingly, for 60 to 64 bit composites, the LP variation runs at almost exactly the same speed as without the LP. For one test value, using a factor base of 60 large primes, the LP matrix had 54 full relations and 14 LP relations. The NonLP variation had 68 full relations. While the LP variation did do less sieving, it was offset by the slightly denser matrix (14 rows were double density), [solved by Gaussian elim) and by the slightly larger amount of arithmetic needed for both combining the LP rows together, and multiplying the relations in a solution set together. I have only tried a small number of examples, but the LP variation seems to run at about the same speed. I have tried varying the sieve line length, varying the factor base size, varying the LP bound. When optimally chosen, both versions of the code run at about the same speed. 

20101004, 20:29  #37 
May 2008
3×5×73 Posts 
If using 3LP on the algebraic side with the ggnfs sievers, add the M 1 parameter to get a little more speed.
This controls which side gets mpqs treatment first. 0 for algebraic and 1 for rational, with 0 being the default. If the first side does not factor nicely, then the other side is skipped. So if you are doing 3LP on the algebraic side, trying mpqs on the rational side first will save the most work if it fails. 
Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Kaby Lake Memory Speed  airsquirrels  Hardware  12  20170622 14:48 
"Hybrid Memory Cube" offers 1 Tb/s memory bandwith at just 1.4 mW/Gb/s  ixfd64  Hardware  4  20111214 21:24 
Different Speed in different OS's  Dubslow  Software  11  20110802 00:04 
Line sieving vs. lattice sieving  JHansen  NFSNET Discussion  9  20100609 19:25 
Combined Sieving speed  Joe O  Prime Sierpinski Project  7  20060922 09:34 