News

[mdettweiler]

Originally Posted by Christenson

I need to ask if you are interested in having CUDA TF available...

FYI, for Riesel numbers (k*2^n-1) and their close cousins the Proth numbers (k*2^n+1), the prefactoring process is a little different than how it's done for Mersennes. Instead of trial factoring each individual candidate to a specific bit level, the most efficient way to do it for these numbers is to use a sieve to screen out factors over a wide range of candidates: for instance, all n<10M at once.

The current state of the art sieving programs for these numbers are tpsieve and the srsieve family of sieves (srsieve, sr1sieve, sr2sieve, and sr5sieve, each being particularly applicable for different scenarios). tpsieve has been ported to CUDA (where a similar speedup over CPUs has been realized, akin to that with mfaktc for Mersenne numbers); it works most efficiently on very large continuous ranges of k and n, and as such it is most well suited to a large project. Currently, the PrimeGrid project is using this program through BOINC to sieve all of k<10000, n<6M on both the Riesel and Proth sides simultaneously; the sieve files produced by this effort are then made freely available to other projects (such as NPLB and RPS in the mersenneforum, and individual searchers coordinating in this subforum). With all the GPU power being thrown at this effort, everything below n=3M is at this point fully sieved to the optimal factor depth (the point at which CPUs can run primality tests faster than the GPUs can find factors); the current range in progress is for n=3M-6M, with n=6M-9M in the early initial stages of sieving.

For some more specialized searches (for instance, such as those done by the Conjectures 'R Us project here at mersenneforum), tpsieve's preference for large swaths of k and n works against it; for these, one needs to use the srsieve programs, which unfortunately have not yet been ported to CUDA. I talked to the developer of tpsieve (Ken_g6 on this forum) about this, and he explained that srsieve's algorithm is much more difficult to implement on a GPU; he thus is not planning to undertake the effort in the near future. If anyone else, however, would like to try it, he would have the everlasting gratitude of the Conjectures 'R Us participants and others doing similar searches.

Hopefully this explains things a bit!

Max

[Christenson]

Terrible job...just terrible....*not!*

It does put a good bound on what to do with mfaktc, though...if I can ever get out from under work....
what wblipp had asked for was an mfaktc-style TF on (41)^(large prime * various small, very smooth composites such as 2^3)-1.

It doesn't sound like it's worth it to extend to reisel or proth numbers.

[wblipp]

Originally Posted by Christenson

what wblipp had asked for was an mfaktc-style TF on (41)^(large prime * various small, very smooth composites such as 2^3)-1.

I hope you are using 41 as a representative small number, not a hard coded constant. I'm interested in this for many small primes, not just 41. I'm worried that I have not accurately conveyed that idea.

[kar_bon]

Updates:

- k-values in page 8000<k<10000 are sorted
- page for RPS Drive #7 completed (some missing countings still there)

[kar_bon]

New page for RPS Drive #11 inserted.

[kar_bon]

I've included a page for the Project "TPS - Twin Prime Search" (under "Other Projects").

Data included (up to 2010-10-19 so far):

- primes found (with person, date)
- number of candidates tested and primes found by user
- distribution of primes (table and graph)
- graph with pairs returned to LLRnet/PRPnet server per day
- ranges overview

Some data from that:
- 159 primes found
- 804016 candidates tested

[kar_bon]

I've included an ASCII-file with Riesel-primes for 10000 < k < 100000 and

- all k's: n<=1007 (from G.Barnes)
- k < 15000: n<=20000 (from T.Ritschel)

Thanks both for the data.

Some numbers:
- the file is ~3MB in size
- 45000 k-values
- 560708 primes
- 15280 twins

The table gives for all k-values the number of primes and the Nash-weight, too.
Twins are marked with '*'.

[kar_bon]

PrimeGrid found (by Timothy D. Winslow) the lagest Twin so far: 3756801695685*2^666669±1 on 2011-12-25.

[Dubslow]

That's my last name. Who the hell is he?

Hmm. I found another Timothy Winslow online (and a whole family of Winslows. Presumably there are many of us.)

[kar_bon]

New Riesel Prime found by PrimeGrid:

162941*2^993718-1 found by D.Domanov.

This prime was overlooked by the RieselSieve-project.

Now 56 candidates left.

[kar_bon]

Next Riesel Prime just verifying:

252191*2^5497878-1 should be place 21 on Top5000.