Sieving discussion thread

[axn]

I am assuming you got this from this page: http://primes.utm.edu/top20/page.php?id=1

If so, the key sentence would be:

Originally Posted by prime pages

Nice--too bad it is of virtually no practical value!

[jasong]

I am assuming you got this from this page: http://primes.utm.edu/top20/page.php?id=1

If so, the key sentence would be:

Didn't people once say the same thing about imaginary numbers?

Anyway, I just thought I'd throw it out there, just in case it was overlooked.

Never mind.

[victor]

FYI, I started sieving 25G-50G. I'm planning to stop when p reaches 3T.

[gribozavr]

I don't think it is worthful. I can do that in 7-10 hours (or so I think) and the datafile will be quite large (> 300Mb) -- and thus not really transferrable.

[victor]

Originally Posted by gribozavr

I don't think it is worthful. I can do that in 7-10 hours (or so I think) and the datafile will be quite large (> 300Mb) -- and thus not really transferrable.

Ok, then I'll keep testing numbers ;)

[biwema]

Splitting up horizontal sieving does not make sense, at all. (Of course, for the first G (before merging) this rule does not apply; here a bitmap instead of a hashtable is used due to the large number of candidates)
The sieving speed does not depend on the size of the range, therefore it is recommended, that the range is initially sized larger then necessary. We just need to take care that the computer has enough memory. (I do also some sieving on a 100G range and I have no impact on the sieving speed. The safe file is 750MBytes (but it can compressed down to 20% if the sieving is distributed) and NewPGen takes 320 Mbytes for the hashtable)

n=195000 does also follow this strategy; they chance of failing is only 1/6. In case we do not find a twin after the first 20G we can discuss our future plans. Either continuing from 25G or choosing a new N is an option. (I suggest to choose a N which is some 100 below a FFT change point (SSE2 and nonSSE2) in order to improve efficiency)

[pacionet]

I think that we'll find a twin prime before 20G.

[smh]

Originally Posted by biwema

n=195000 does also follow this strategy; they chance of failing is only 1/6. In case we do not find a twin after the first 20G we can discuss our future plans. Either continuing from 25G or choosing a new N is an option. (I suggest to choose a N which is some 100 below a FFT change point (SSE2 and nonSSE2) in order to improve efficiency)

I know it's a bit early, but shouldn't we start thinking about the next N already?

[gribozavr]

Range [1200e6; 25e9] is at p=930.0T, 8,397,809 k's left.

[biwema]

Originally Posted by smh

I know it's a bit early, but shouldn't we start thinking about the next N already?

Some time ago I did some testing and continued quite a bit.

My favourite is N=333333.
It has a little bit more than 100000 Digits and is just below a FFT change for SSE2.
So, I already started sieving on that number.
I have chosen a range of 100G (Has the same probability of finding a twin like a 34G Range at 195000).
At the moment I am somewhere beyond 150T and have less than 39 million candidtes left.

I will continue sieving that range and we can start with that N after finishing. N=195000. Maybe even distributes sieving is possible; the ideal sieving depth is between 50P and 100P depending architecture used.

How do you think about it?

Edit: Imagaine, we are working on than N, it is very likely that we find thousands of primes that are not a twin. These Primes will flood the top5000 page and have quite an inpact on the diversity of primes.

biwema

[gribozavr]

Looking at http://primes.utm.edu/primes/lists/all.txt I see that 17 primes have already been reported for n=333333.