Even after reading only the first 3 paras. in Curtis's post, I have to defer to him here. (I'll read the rest in a min.) He has done an extensive study of this stuff. Even assuming that sieving far deeper as he suggests is correct, it causes the lower n-range to be sieved so much farther then the optimum had you sieved a smaller n-range. Personally, I would be too impatient.

In other words, I want to sieve the lower n-ranges to a depth that would be optimum had I been sieving, say, n=50K-100K or n=50K-150K instead. That way, I'm not waiting for what seems like forever to start such a low n-range.

I was wondering if this thread would bring out the math guys. It did! lol

Edit: After reading all of Curtis's post: My response: Do that! (That is, like he said, if you definitely plan on testing all of n=50K-250K and you can handle waiting to get started.) If you want to sieve the whole range but only test n=50K-100K or n=50K-150K and leave the rest for others, I'd suggest the method that I suggested modified with your fftlen analysis. The 70% of the range suggested by me and 1/3rd of the high range suggested by Masser are invalid points if you are breaking it off at every fftlen change where the testing time change within each 'piece' is strictly linear.


Gary

[quote=masser;160371]Minor quibble - it's not an exponential increase, but a polynomial increase.[/quote]


I couldn't think of the term; hence the "i.e." to clarify what I meant. Thanks for the clarification.

How do you estimate the optimal sieve depth? For example I need an elimination of about 1500 - 1750 sec/n before I want to break off a piece. I'm currently sieving at 670G doing about 900s/n for 52180 remaining candidates.
How do I start calculating this?