View Single Post
Old 2020-10-15, 14:26   #2169
Quasi Admin Thing
KEP's Avatar
May 2005

953 Posts

Originally Posted by VBCurtis View Post
I'm playing with srbsieve sieve & phase settings, so I may find some more speed; I'm trying 5 phases this time rather than 4 last time, for instance. 15 is far too many! Also, I sieve deeper than 78e6 on the last two phases; perhaps I'll try a bit less sieving to see if overall speed improves.
First up, please tell me that you did only 6 sieves, for the entire k range, using NewPGen? ... you ARE WASTING ressources if you sieve only 1M k's and then continue sieve the other ranges. SRBsieve does not care if the amount of k's extent beyond the maxK value in the srbsieve.ini file. It sounds reasonable to split your range in multiple million k ranges - however it may as you may have noticed, affect the sievevalue.

There may be room for optimizations, when it comes to the .ini files. The values should, if one decides not to remove any phases, be the most (likely to be) optimal. I did a lot of calculations over the 6 months I worked on these .ini files, most of the calculations were to have optimal sievevalues for each phase for b=1030 and then scale towards any other base. It took a lot of testing but the result was the .ini files, where the sievevalue was determined by amount of tests remaining at a fixed p value and time per test for given base and FFT value.

If you find optimizations, it may be time someone redo the .ini files for the amount of bases remaining untested.
KEP is offline   Reply With Quote