Team sieve #22: c166 from 3270:620 - Page 5

bsquared · 2011-01-31, 23:19

54-56M done: 5461598 relations.

upload will begin shortly...

fivemack · 2011-02-02, 09:31

After a small delay because I wanted to fill in the gaps, I've started the linear algebra.

Code:

Wed Feb  2 05:06:18 2011  found 25188765 hash collisions in 120443951 relations
Wed Feb  2 05:07:50 2011  commencing duplicate removal, pass 2
Wed Feb  2 05:18:15 2011  found 27239691 duplicates and 93204260 unique relations

Wed Feb  2 07:04:48 2011  matrix is 9193549 x 9193773 (2706.7 MB) with weight 698835592 (76.01/col)
Wed Feb  2 07:04:49 2011  sparse part has weight 617599344 (67.18/col)
Wed Feb  2 07:04:49 2011  using block size 65536 for processor cache size 6144 kB
Wed Feb  2 07:05:37 2011  commencing Lanczos iteration (4 threads)
Wed Feb  2 07:05:37 2011  memory use: 2388.5 MB
Wed Feb  2 07:08:53 2011  linear algebra at 0.0%, ETA 316h13m

11085 nfsslave  20   0 3228m 3.1g 1112 R  400  9.9 678:28.33 msieve

Should get factors about Feb 16th. This is running four threads on an eight-core dual-quad-Opteron, because in the past I've found that eight threads ran more slowly than four.

Batalov · 2011-02-02, 09:48

For the Opteron, you may want to try a -DLARGE_BLOCKS binary . . .

fivemack · 2011-02-15, 17:27

Bother.

Thank to everyone for their sieving help, but this turns out to have been a horrible ECM miss: I did do 12000 curves at B1=1e7, which was about 600 CPU-hours, but failed to find

Code:

Tue Feb 15 16:58:26 2011  prp50 factor: 12769918819254763810340639239833146584153360860151
Tue Feb 15 16:58:26 2011  prp117 factor: 138863197118762133922694063802006943284767706883225766187196880289575580789111545331577944293745560124154160404025071

which appeared after sieving effort equivalent to ~9000 CPU-hours on the same hardware

FactorEyes · 2011-02-15, 17:53

Quote:

Originally Posted by fivemack

I did do 12000 curves at B1=1e7, which was about 600 CPU-hours ...

I assume you mean B1=11e7, which would have put the chance of missing a p50 at around 12%.

The Gods did not favor this particular endeavor.

fivemack · 2011-02-15, 19:43

No, I meant 1e7; over-optimism on my part. Indeed 12000@1e7 probably doesn't find a P50, but I should have run something nearer 12000@3e7 which would have found it with reasonable probability.

bsquared · 2011-02-15, 19:52

12000 curves at 11e6 (my script doesn't include 10e6 as a default) has a probability of missing a p50 of 70%. Estimating each 10e6 curve as 10/11ths of a 11e6 curve gives a probably of missing a p50 of 72%. Either way, its a fairly high miss probability. And either way, I'm still happy to have contributed. No regrets

fivemack · 2011-02-15, 20:18

May I ask who's running on with this sequence? I had one core of a Q6600 working on it, but noticed that the results in the DB had got further than I had (extracting several P3x factors)

I don't mind if someone else is running it - I have about a dozen other sequences checked out with Christophe - but please say who you are and say here when you stop.

evandijk70 · 2011-02-15, 21:11

Sorry, I thought it was a free for all in the first few hours after a factor had been found, like in the other team sequence. I had a few Q6600 cores running on it as well. I won't be able to run it any further though (past the c122). It had 800 curves @3e6 and 100 @11e6.

Andi47 · 2011-02-15, 21:28

A c122 should be quite possible with "a few Q6600 cores". I have done c125's (and a bit higher) with a C2D@1.8 GHz (I guess it's an E4300)

fivemack · 2011-02-15, 22:26

Well, yes, it will take 44 hours on one core of Q6600 - not big enough to run on more than one core, really.

I'll continue running the sequence now (IE have started gnfs polsel on the C122)

2011-02-15, 17:27	#48
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 7²×131 Posts	Bother. Thank to everyone for their sieving help, but this turns out to have been a horrible ECM miss: I did do 12000 curves at B1=1e7, which was about 600 CPU-hours, but failed to find Code: Tue Feb 15 16:58:26 2011 prp50 factor: 12769918819254763810340639239833146584153360860151 Tue Feb 15 16:58:26 2011 prp117 factor: 138863197118762133922694063802006943284767706883225766187196880289575580789111545331577944293745560124154160404025071 which appeared after sieving effort equivalent to ~9000 CPU-hours on the same hardware

2011-02-15, 20:18	#52
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 6419₁₀ Posts	May I ask who's running on with this sequence? I had one core of a Q6600 working on it, but noticed that the results in the DB had got further than I had (extracting several P3x factors) I don't mind if someone else is running it - I have about a dozen other sequences checked out with Christophe - but please say who you are and say here when you stop. Last fiddled with by fivemack on 2011-02-15 at 20:19

2011-02-15, 21:11	#53
evandijk70 Nov 2009 5 Posts	Sorry, I thought it was a free for all in the first few hours after a factor had been found, like in the other team sequence. I had a few Q6600 cores running on it as well. I won't be able to run it any further though (past the c122). It had 800 curves @3e6 and 100 @11e6. Last fiddled with by evandijk70 on 2011-02-15 at 21:12

2011-02-15, 22:26	#55
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 1100100010011₂ Posts	Well, yes, it will take 44 hours on one core of Q6600 - not big enough to run on more than one core, really. I'll continue running the sequence now (IE have started gnfs polsel on the C122) Last fiddled with by fivemack on 2011-02-15 at 22:27

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Team sieve #26: c166 from 4788:2661	jrk	Aliquot Sequences	38	2011-05-16 17:58
Team sieve #25: c166 from 4788:2632	jrk	Aliquot Sequences	13	2011-03-24 02:48
Team Sieve for 2995125705	SlashDude	Riesel Prime Search	78	2006-05-14 16:56
Team Sieve	grobie	Riesel Prime Search	3	2005-11-16 08:46
Team Sieve of 210885	SlashDude	15k Search	21	2003-12-23 16:31

2011-01-31, 23:19	#45
bsquared "Ben" Feb 2007 3×1,171 Posts	54-56M done: 5461598 relations. upload will begin shortly...

2011-02-02, 09:48	#47
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 22405₈ Posts	For the Opteron, you may want to try a -DLARGE_BLOCKS binary . . .

2011-02-15, 19:43	#50
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 14423₈ Posts	No, I meant 1e7; over-optimism on my part. Indeed 12000@1e7 probably doesn't find a P50, but I should have run something nearer 12000@3e7 which would have found it with reasonable probability.

2011-02-15, 19:52	#51
bsquared "Ben" Feb 2007 3×1,171 Posts	12000 curves at 11e6 (my script doesn't include 10e6 as a default) has a probability of missing a p50 of 70%. Estimating each 10e6 curve as 10/11ths of a 11e6 curve gives a probably of missing a p50 of 72%. Either way, its a fairly high miss probability. And either way, I'm still happy to have contributed. No regrets

2011-02-15, 21:28	#54
Andi47 Oct 2004 Austria 2×17×73 Posts	A c122 should be quite possible with "a few Q6600 cores". I have done c125's (and a bit higher) with a C2D@1.8 GHz (I guess it's an E4300)