Sieve Discussion - Page 6

em99010pepe · 2007-05-17, 18:31

To Geoff.

I removed the latest factors found from all attack sides such sieving and P-1 factoring and got a file with 2897 terms.
How many terms do you have on yours?

Carlos

geoff · 2007-05-17, 23:02

Quote:

Originally Posted by em99010pepe

I removed the latest factors found from all attack sides such sieving and P-1 factoring and got a file with 2897 terms.
How many terms do you have on yours?

I had just removed the ones found by sieve, 2917 remain. I'll use your sieve file instead, the fewer terms the better :-)

em99010pepe · 2007-05-19, 08:54

After completing the 2800k-2900k range at P-1 and almost finishing my sieve range, I updated my sieve file. 2889 terms left.

hhh · 2007-05-30, 08:16

All right, here we go.

Sieving the first stage is finished, and certainly our attention should be pointed at LLR and the remaining P-1 ranges to prepare it.

Nevertheless, the sieving part of the next stage is going to take a longer time (given the increased worth of every factor), and has already begun. I'm sieving right now the range 5M-25M, and the final number of tests to LLR will probably be around 28000. But, how will we proceed?

I suggest that I keep going, crunching what I can on one machine. This would have the advantage of minimising the risk of human error in the factor transmission.
Then, when LLR is approaching 4M, we could make the sieving public, and push it together a bit further, say up to 5G, for the whole range 5M-25M.
And finally, when LLR approaches 5M and we are going to actually need the numbers, we could split the range into slices of 1 M and sieve them as the need advances.

What do you think? Better ideas? Is there anything to poll?

Yours H.

geoff · 2007-06-05, 02:02

The FFT sizes used by LLR for the 5-10 million range will be 512k, 610k, 768k, 896k, 1026k. Instead of reducing the sieve by removing the lowest 1 million n range, we could reduce it just to exclude the smallest FFT size.

hhh · 2007-06-06, 21:16

That's a good idea. Can you determine and post the change levels? Sieving to 100G is going to be finished around the 21st of June, we'll see then how to proceed. Yours H.

geoff · 2007-06-07, 22:23

Here are the approximate break points between FFT sizes for SSE2 machines:

Code:

FFT size   Break point
--------   ------------------
448/512    4522000 / 4523000
512/640    5126000 / 5127000
640/768    6423000 / 6424000
768/896    7652000 / 7653000
896/1024   8912000 / 8913000

If sieving for the 5-10 million range reaches the high point for the 1.5-5 million range (4200G) well before LLR reaches the 448/512 breakpoint, then maybe it would be worthwhile including the full 512k FFT range, i.e. sieve 4.5-10 million.

2007-06-07, 22:23	#62
geoff Mar 2003 New Zealand 1157₁₀ Posts	Here are the approximate break points between FFT sizes for SSE2 machines: Code: FFT size Break point -------- ------------------ 448/512 4522000 / 4523000 512/640 5126000 / 5127000 640/768 6423000 / 6424000 768/896 7652000 / 7653000 896/1024 8912000 / 8913000 If sieving for the 5-10 million range reaches the high point for the 1.5-5 million range (4200G) well before LLR reaches the 448/512 breakpoint, then maybe it would be worthwhile including the full 512k FFT range, i.e. sieve 4.5-10 million.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
SIEVE GAP	pepi37	Other Mathematical Topics	2	2016-03-19 06:55
Advantage of lattice sieve over line sieve	binu	Factoring	3	2013-04-13 16:32
Combined Sieve Guide Discussion	Joe O	Prime Sierpinski Project	35	2006-09-01 13:44
Sieve discussion Meaning of first/second pass, combined	Citrix	Prime Sierpinski Project	14	2005-12-31 19:39
New Sieve Thread Discussion	Citrix	Prime Sierpinski Project	15	2005-08-29 13:56

2007-05-30, 08:16	#59
hhh Jun 2005 565₈ Posts	All right, here we go. Sieving the first stage is finished, and certainly our attention should be pointed at LLR and the remaining P-1 ranges to prepare it. Nevertheless, the sieving part of the next stage is going to take a longer time (given the increased worth of every factor), and has already begun. I'm sieving right now the range 5M-25M, and the final number of tests to LLR will probably be around 28000. But, how will we proceed? I suggest that I keep going, crunching what I can on one machine. This would have the advantage of minimising the risk of human error in the factor transmission. Then, when LLR is approaching 4M, we could make the sieving public, and push it together a bit further, say up to 5G, for the whole range 5M-25M. And finally, when LLR approaches 5M and we are going to actually need the numbers, we could split the range into slices of 1 M and sieve them as the need advances. What do you think? Better ideas? Is there anything to poll? Yours H.

2007-06-05, 02:02	#60
geoff Mar 2003 New Zealand 10010000101₂ Posts	The FFT sizes used by LLR for the 5-10 million range will be 512k, 610k, 768k, 896k, 1026k. Instead of reducing the sieve by removing the lowest 1 million n range, we could reduce it just to exclude the smallest FFT size.

2007-06-06, 21:16	#61
hhh Jun 2005 373 Posts	That's a good idea. Can you determine and post the change levels? Sieving to 100G is going to be finished around the 21st of June, we'll see then how to proceed. Yours H.