Sieving discussion thread

[Cybertronic]

/ 6 is a good hint. I get 57MB. All differences encode in dual numbers and then
compress,but the zip-file is 67MB, not<57 :-(

[Cybertronic]

Originally Posted by Cybertronic

/ 6 is a good hint. I get 57MB. All differences encode in dual numbers and then
compress,but the zip-file is 67MB, not<57 :-(

Latest and best compression from me is 1:12,7: File is 54.8 MB (old 697MB), but is not relevant 55 or 65 MB I mean.

[jasong]

Possibly stupid question, but...

Is it possible to make a siever that sieves on more than one core, but have only one instance of the list in RAM? I'm thinking there could be multiple threads finding potential factors, but only one thread testing them.

[jmblazek]

Originally Posted by MooooMoo

Anyway, I've started sieving 50G-208G. My progress is at 26T. Unless something unexpected happens, it'll be close to (or at) the optimal sieving depth after n=333333 is complete, and after 1-50G is done for n=500000.

The optimal sieving depth, as provided by biwema, for n=500000 is 2^57.5 which is 203P. On a single P4 3.6 GHz machine, it's estimated to take over 40 years.

Either you have a GREAT PC farm or the twin for n=333333 is going to take a really long time. I hope it's the first!

By "optimal" sieving depth I presume you mean somewhere around 2^50...or 350 k per M??? in which case, well within reach in a year at 3T a day. Although I hope it doesn't take us a year to find the next twin.

p.s. And by the time n=500000 delivers a twin, maybe 1 G Ram will be $40.

[MooMoo2]

Originally Posted by jmblazek

The optimal sieving depth, as provided by biwema, for n=500000 is 2^57.5 which is 203P. On a single P4 3.6 GHz machine, it's estimated to take over 40 years.

Either you have a GREAT PC farm or the twin for n=333333 is going to take a really long time. I hope it's the first!

By "optimal" sieving depth I presume you mean somewhere around 2^50...or 350 k per M??? in which case, well within reach in a year at 3T a day. Although I hope it doesn't take us a year to find the next twin.

p.s. And by the time n=500000 delivers a twin, maybe 1 G Ram will be $40.

I misread the optimal sieving depth (thought it was 20.3P, not 203P). My PC farm isn't that great either Just 2 Pentium 4's, 1 Pentium 3, and 1 Celeron (900 MHz). Other than the 10T I sieved earlier, I'm not participating in any TPS/PrimeGrid efforts for n=333,333, so that I can put as much computing power as posssible into n=500,000.

If the twin is found really early (next month :surprised ), my plan is to ask pacionet to release 1-50G first. By the time PrimeGrid finishes with 1-50G, my range (50G-208G) will be at 20.3P.

edit: I'm averaging around 40T per day. With the addition of a Core 2 Quad sometime this spring, production should double, which would make 20300T (20.3P) achievable by September - October 2007.

[biwema]

Originally Posted by pacionet

Ok I'll be able to sieve, for the beginning, the range 0-50G for exponents around n=500,000.

As soon as we decide n I'll start with that range.

Suggestion: Split up the range in two pieces *at the beginning*

When sieving, the starting bitmaps are merged at the 1G threshhold. At that level my 100G had 1.7GByte. That was no problem for me.

If you sieve 300G (So we have even a better chance to find a twin (100G at 333333 equals 250G at 500000)), The file is more than 5GByte at 1G.

I recomend to split is up the sieving in two pieces:
0-150G and 150G -300G (or 3 pieces if only files up to 2G are allowed)

Below 1T the size of the file is dropping very quickly. At 1T (less than 1 day of sieving) each of the piece has maybe 1.5 GByte. At this level you can connect the savefiles (remove the headerline of the second one and concat them). So you get a file which is around 3GByte. This one you can sieve without getting errors.
After sieving to 500T or less, the whole file is even smaller than 2G, which can be stored older harddirves.

[pacionet]

Originally Posted by MooooMoo

I misread the optimal sieving depth (thought it was 20.3P, not 203P). My PC farm isn't that great either Just 2 Pentium 4's, 1 Pentium 3, and 1 Celeron (900 MHz). Other than the 10T I sieved earlier, I'm not participating in any TPS/PrimeGrid efforts for n=333,333, so that I can put as much computing power as posssible into n=500,000.

If the twin is found really early (next month :surprised ), my plan is to ask pacionet to release 1-50G first. By the time PrimeGrid finishes with 1-50G, my range (50G-207G) will be at 20.3P.

edit: I'm averaging around 40T per day. With the addition of a Core 2 Quad sometime this spring, production should double, which would make 20300T (20.3P) achievable by September - October 2007.

I did the same bad computations; moreover I have some problems now on my PCs.
If anybody else (who has sufficient computing power,...) wants to sieve 0-50G replacing me, tell me.

Otherwise I'll sieve 0-50 but slowly.

[pacionet]

KEP please tell me as soon as possible if you want 0-50G range otherwise I'll continue sieving.

Thanks

[philmoore]

I see that David Broadhurst is suggesting at
http://tech.groups.yahoo.com/group/primeform/ ,
see especially
http://tech.groups.yahoo.com/group/p...m/message/8322 ,
that by sieving on both prime twins and Sophie Germain candidates, you have the potential for finding not just one but two new records.

A Sophie Germain pair p, 2p+1, where p = 3 mod 4, also gives rise to a Mersenne number 2^p-1 with factor 2p+1. As it is, I believe that the largest Mersenne number with prime exponent known to be composite corresponds to the largest known Sophie Germain pair. (Someone please correct me if I am wrong.)

[MooMoo2]

Originally Posted by philmoore

I see that David Broadhurst is suggesting at
http://tech.groups.yahoo.com/group/primeform/ ,
see especially
http://tech.groups.yahoo.com/group/p...m/message/8322 ,
that by sieving on both prime twins and Sophie Germain candidates, you have the potential for finding not just one but two new records.

A Sophie Germain pair p, 2p+1, where p = 3 mod 4, also gives rise to a Mersenne number 2^p-1 with factor 2p+1. As it is, I believe that the largest Mersenne number with prime exponent known to be composite corresponds to the largest known Sophie Germain pair. (Someone please correct me if I am wrong.)

I've heard of sieving on both types of candidates. You're not the first (or second) person to suggest it.

This will probably (in fact, 99.99% likely) not happen, because sieving on n=333,333 has already started. I don't like the idea of throwing away over 1P worth of sieve work. n=500,000 hasn't gone that far yet, but I can't get access to three of my comps that are doing the sieving until the start of spring break (they're in my parent's house, and I'm in college now). By that time, sieving will likely be above the 1P level too.

I don't know what the n after 500,000 will be, but don't hold your breath. It's likely that the range of k needed for sieving will be so large that it will be over the range at what LLR and NewPGen can reliably test. I remember reading on another forum that LLR slows down and is not fully reliable past k values larger than 2^52 or so (I forgot the exact limit, but it's somewhere around there).

P.S. A random rant
I just read a comment in there that states "avoidance of silly MooMoo names." Hmmm. Oh well, I'm still keeping it the same

And my Barney avatar isn't coming off either

[ValerieVonck]

Hi everyone!

I have small question: if I want to hunt for a TWIP myself what type do I choose in NewPGen?

- Twin?
- SG
- CC
- BiTwin
- Twin/SG
- Twin/CC

Thank you

Ps. Should my input look like

base=2
n=450000
kmin=2
kmax=100.000.000.000