Sieve (for advanced users)
 

Any one interested in increasing their stats and helping test some code/ideas?

Download non combined .dat file from [url]www.psp-project.de[/url]
Download srsieve from here
[url]http://www.geocities.com/g_w_reynolds/sr2sieve/testing/[/url]

Use command line
sr2sieve -y 6 -s --pmin 800000000000000 --pmax 801000000000000
(change ranges on what values you reserve. This sieves from 800T to 801T)

800T onwards available
Please reserve range in intervals of Trillions.

Post reservations below.
(This thread does not change the other sieve thread, we will still have to do the ranges in the other thread.)

I tested some stuff and am finding a factor every 6 hrs. How long does it take you to find a factor?

800T+ available.
:smile:

800T -> 801T Reserved CedricVonck :smile:

Already found 1 factor (4 hours)
I am running @ 21M / sec (P4 2800 - 1Gb Ram)

Citrix please start with these ranges... not all of the ranges above 800T have been sieve with the 991-50M dat for SoB.

These are the ranges which have been sieve with the 99-150M dat and only require use of the psp-only dat.

805500 806500
809500 811500
811500 811600
811600 812700
818000 820400
820400 820500
825500 828500
829000 831000
831000 831500
831500 832000
833000 833300
833800 840000

VJS,

We are just doing a few Trillions for right now to see the usefulness of algorithm. Nothing major. Once Geoff implements the cutoff model I will take your request into account. (Also hhh and ltd will be back to handle this thread better.)

The major use of using this algorithm is to save PRP tests. (As eventually we will have to sieve the whole range over anyway). So I was thinking of just sieving from n=4M to 7.5M. Is there a range that SOB needs sieving in? What level are they PRPing at?


:smile:

Citrix,

Your on the same wave length as myself, I tried the implementation with the 7 k remaining with the 991-50M dat. On a 2.2G Barton core I was getting 50M/sec.

With this result I was thinking about trying a very small n-range.

Perhaps 13.4M to 15M with the remaining 7k SoB k at a very high p around 2000T just too see what happens. I am hoping to see in an equivalent in crease in speed with respect to an additional decrese in n-range.

My only concern is that the current effort would be wasted since the range would need to be resieved later. The corrolary is that the sieve could become ineffective as a result and we would never sieve the range fully...

Not sure how I feel about it from a project user morality basis...

After this I will have to compare the CPU time invested vs the return in factors. If anything this may be a replacement for p-1 efforts.

If people could post their results with p and factors for k/n pairs in this thread it would be helpful.

VJS, I think we will use this as a replacement for P-1. I don't think it will help the main sieve as we will have to sieve the non-smooth values anyway.

Have you tried -y 12 setting. Seems very fast.:smile:

Range:

800T -> 801T Complete

Factors found (expected 50+)

[code]
800070176564161 | 265711*2^35785320+1
800604447457921 | 225931*2^29830952+1
[/code]

Sr2sieve log

[code]
05/22/07 12:31:25 WARNING: Only testing for factors p of the form p=1 (mod 64).
05/22/07 12:31:25 sr2sieve started: 2 <= n <= 49999947, 800000000000000 <= p <= 801000000000000
05/22/07 15:59:47 sr2sieve stopped: at p=800259314379457 because SIGINT was received.
05/22/07 15:59:47 Found factors for 1 term in 12194.635 cpu sec. (expected about 12.91)
05/23/07 08:48:10 WARNING: Only testing for factors p of the form p=1 (mod 64).
05/23/07 08:48:11 sr2sieve started: 2 <= n <= 49999947, 800259314379457 <= p <= 801000000000000
05/23/07 15:48:43 sr2sieve stopped: at p=800774037059137 because SIGINT was received.
05/23/07 15:48:43 Found factors for 2 terms in 36513.313 cpu sec. (expected about 38.51)
05/24/07 10:43:50 WARNING: Only testing for factors p of the form p=1 (mod 64).
05/24/07 10:43:50 sr2sieve started: 2 <= n <= 49999947, 800774037059137 <= p <= 801000000000000
05/24/07 13:51:41 sr2sieve stopped: at p=801000000000000 because range is complete.
05/24/07 13:51:41 Found factors for 2 terms in 47287.395 cpu sec. (expected about 49.75)
[/code]


-y 12 yields:

p=800081339047937, 1.372.926.103 /sec on a P4 2800Ghz 1 Gb Ram

Cedric,

So what your saying is that the estimation of the number of factors found is not close to reality???

The expected # of factors is assuming you are using y=1. Since we are using y=6, we should find 2^5 times less factors.

50/32=~2. Which is what we found.
Time per factor =6.5 hrs
So not bad.

[QUOTE=Citrix;106698]Any one interested in increasing their stats and helping test some code/ideas?
[/QUOTE]

If you're interested in some non-x86 (PPC64/Linux) testing, I'd be happy to donate some cycles. I've got at least 4 different variants ppc970 variants. All I need is the source which includes the -y flag functionality.

:alex: