Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummerslowrise

11)
Message boards :
News :
World Record Generalized Woodall Prime
(Message 107778)
Posted 1496 days ago by serge
Glad to see the announcement from John, and such early results from the project. Great work everyone!
Yes, sometimes the first news post here is old news. Hopefully people do not misunderstand this for a 2017 world record GW prime. /JeppeSN
There isn't though. Is there?
For all I see, there is a 2015 World Record Generalized Woodall Prime, and it belongs to ...hmmm... look at that, who could have guessed?
Of course not for long now, but still.

12)
Message boards :
General discussion :
Your Erdos Number
(Message 101671)
Posted 1653 days ago by serge
Mine is 3.
Batalov > Lander > Kleitman > Erdos

13)
Message boards :
Number crunching :
Disturbance
(Message 100647)
Posted 1680 days ago by serge
The Fermat divisor test will take a month and a half on a Haswell.
It doesn't really take a month and a half  less than a week  and that's even on nonparallel code, and not on a Haswell.

14)
Message boards :
Sophie Germain Prime Search :
Sophie Germain Prime Search
(Message 92565)
Posted 1930 days ago by serge
I have not checked how TwinGen is coded (and more importantly how was it actually applied), but NewPgen (see source here) proceeds by iterating over a prime generator, finding the modular exponent 2^n mod p (or (1/2)^n mod p for some sieving modes) and then removing k values for all requested composite candidates for each of the four variations (there is no room for "OR"s in that algorithm; if any of the four derivative forms k*2^(n+x)+c (where x and c are some pertinent subset of 1,0,1, ...as well as some higher x>1 for CC sieving) is divisible by p, then k is struck out; the bitmask/sparsearray structure for all prior k values does not have any counters of strikeouts or similar  only 1 or 0 for each k). NewPgen has some excellent heuristics for managing the large bit arrays and switches between when the initially huge input range collapses, but it does manage single bits; no second chances (say, 3 out of 4) are given  the latter can only be arranged by external scripting; it's not hard even if a bit tedious.
TwinGen page is quite informative on multisieve principles, and (I quote) "The decision criteria and discussion is more than can be easily typed into an introductory web page."

15)
Message boards :
Sophie Germain Prime Search :
Sophie Germain Prime Search
(Message 92559)
Posted 1930 days ago by serge
SG pair is a Cunningham chain of length n=2.
How does one maximize their chances of success of finding one? Anyone who pondered about this question for more than a moment would realize that one would want to sieve for chains of length n+1, then check the middle survivor (because this is a computationally expensive check) and then have double chances of completing the chain.
For larger n values, one can improve slightly more; e.g. for a chain of length n=6, one can sieve for length n+2, then check the middle 4 elements {c(i),c(i+1),c(i+2),c(i+3)} and if successful, proceed to checking c(i1),c(i+4); if one of them is prime use the last chance, and if both are prime, one already has a CC6 and still a minuscule chance of CC7 and CC8.
So, notwithstanding that this SG pair was found (congratulations!) and that it would not have been found if quad sieving was done correctly,  the fact is that over the same time/effort you would have found perhaps two SG pairs (different from this one, of course) and/or would have found the first of the two already a year ago.

16)
Message boards :
Number crunching :
test
(Message 91840)
Posted 1958 days ago by serge
/deleted/

17)
Message boards :
Generalized Fermat Prime Search :
GFN Status by nRange (1722)
(Message 91648)
Posted 1963 days ago by serge
Thanks! Looks good (whether it happened automatically or manually).
79K looks like a good buffer now.

18)
Message boards :
Generalized Fermat Prime Search :
GFN Status by nRange (1722)
(Message 91640)
Posted 1964 days ago by serge
GFN_n=17_low task stack is running low, can run out in a couple days.
Generalized Fermat Prime Search (n=17 low) 995 / 8955
Last number used to be >30K recently, but was steadily trickling down.
(As per estimate in another thread, we'd want to test ~36K per one hit. Maybe 4050K candidates could be turned into tasks now?)
Thx.

19)
Message boards :
Seventeen or Bust :
2912 hours CPU for SOB task?
(Message 91516)
Posted 1967 days ago by serge
Winsorisation

20)
Message boards :
General discussion :
Great Internet Mersenne Prime Search New Prime?
(Message 91131)
Posted 1980 days ago by serge
I think the second thing will be the way to go. Neither the speed nor the chances will change (not significantly anyway). The goal of WR is (and always has been, in my opinion) a respectable one.
In fact, I'd encourage to rereview and implement the Cyclo N=22 sieving and an exactly equivalent queue, in parallel. The runtimes will be the same as for an equally sized GFN; the chances of a Cyclo WR are in fact just a bit higher per unit of time! (Because they have less allowable prime factors while sieving and can be sieved higher using the same time.)

Next 10 posts
