Holy Speedup, Batman!

R.D. Silverman · 2008-10-01, 13:47

The LA for 2,823+ and 2,823- seems to have gone so blazingly
fast that I am astonished. The LA for 7,319- was started long before
these two and it is still going! Indeed, the sieving for 2,823+ went
so fast that I was not even aware that it had been started when I saw
it appear on page 109!!!

Kudos!!!!

How did NFSNET get these two done so quickly???????

What's next?

10,247+ is about 85% sieved. I will then do 2,836+.

frmky · 2008-10-01, 18:30

Bruce normally commits only a small portion of his resources to the NFSNet sieving. For 2,823+/-, he used the big Condor cluster and blew through the sieving for each in about 2 days. They ended up significantly oversieved, so the LA was easier than usual. The smaller matrices fit in 4GB, so Bruce was able to complete the LA for both on a faster 4GB quad-core computer.

7,319- used a quintic, so the sieving was harder. We stopped perhaps a bit early, and ended up with a larger matrix running on a slower computer. It has about another 3 weeks to go.

We've been a bit disorganized of late, so we're not yet sure what will come after 10,287-. Perhaps 7,319+.

Greg

fivemack · 2008-10-01, 22:30

I did the linalg for 2-821 a few months back; significant oversieving meant the matrix was

Sat Jun 7 13:07:34 2008 found 84366768 hash collisions in 266645965 relations
Sat Jun 7 13:07:34 2008 commencing duplicate removal, pass 2
Sat Jun 7 13:44:41 2008 found 52049651 duplicates and 214596314 unique relations
Sun Jun 8 17:52:47 2008 matrix is 10129794 x 10130042 (2798.1 MB) with weight 690519820 (68.17/col)

and took 375 hours on a 4GB quad-core. So I'm not particularly surprised that 2^823 \pm 1 didn't take all that long ... the way msieve does filtering makes oversieving decidedly profitable.

R.D. Silverman · 2008-10-01, 22:48

Quote:

Originally Posted by frmky

Bruce normally commits only a small portion of his resources to the NFSNet sieving. For 2,823+/-, he used the big Condor cluster and blew through the sieving for each in about 2 days. They ended up significantly oversieved, so the LA was easier than usual. The smaller matrices fit in 4GB, so Bruce was able to complete the LA for both on a faster 4GB quad-core computer.

7,319- used a quintic, so the sieving was harder. We stopped perhaps a bit early, and ended up with a larger matrix running on a slower computer. It has about another 3 weeks to go.

We've been a bit disorganized of late, so we're not yet sure what will come after 10,287-. Perhaps 7,319+.

Greg

I am truly confused. [i.e. my normal state]

I thought that you and Bruce were a separate group from NFSNET. Yes?
No? If you and Bruce are doing 7,319+, then what will NFSNET do?

I have only a small group of cores. On any given night I have about
15 machines (a max of 22, but some get disconnected, turned off, etc).
Some of these are dual-core. But they are only available weekends
and weeknights. b^n ~ 230 digits (perhaps a little more) is my reasonable
limit. I like to work on an entry in the 'first 5 holes' when one is in reach.
Some of the earlier first holes are too easy for you & Bruce, but someone
needs to do them. I thought that NFSNET was doing them.

Wacky · 2008-10-02, 01:28

Bob,

Greg and Bruce are, and for some time have been, MAJOR contributors to the "NFSNet effort", when they choose to be. They have also done a number of significant projects independently, or in collaboration, outside of the NFSNet framework.

I think that we all have fundamentally non-conflicting goals and that a "conference" related to the independent efforts would be appropriate.

2008-10-01, 13:47	#1
R.D. Silverman Nov 2003 2²×5×373 Posts	Holy Speedup, Batman! The LA for 2,823+ and 2,823- seems to have gone so blazingly fast that I am astonished. The LA for 7,319- was started long before these two and it is still going! Indeed, the sieving for 2,823+ went so fast that I was not even aware that it had been started when I saw it appear on page 109!!! Kudos!!!! How did NFSNET get these two done so quickly??????? What's next? 10,247+ is about 85% sieved. I will then do 2,836+.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
A Holy New Board Game	MattcAnderson	Puzzles	2	2015-10-18 09:40
Holy new Mersenne prime, Batman! (M47 related)	ixfd64	News	901	2008-11-10 19:29
Holy Beaverpotamus, Batman!	ewmayer	Science & Technology	4	2008-03-14 19:19
Holy <censored>! Wow!!!!!!!!!	R.D. Silverman	Factoring	11	2005-04-07 17:00
More RAM = Speedup??	dave_0273	Hardware	8	2004-06-23 05:15

2008-10-01, 18:30	#2
frmky Jul 2003 So Cal 2082₁₀ Posts	Bruce normally commits only a small portion of his resources to the NFSNet sieving. For 2,823+/-, he used the big Condor cluster and blew through the sieving for each in about 2 days. They ended up significantly oversieved, so the LA was easier than usual. The smaller matrices fit in 4GB, so Bruce was able to complete the LA for both on a faster 4GB quad-core computer. 7,319- used a quintic, so the sieving was harder. We stopped perhaps a bit early, and ended up with a larger matrix running on a slower computer. It has about another 3 weeks to go. We've been a bit disorganized of late, so we're not yet sure what will come after 10,287-. Perhaps 7,319+. Greg

2008-10-01, 22:30	#3
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 13×491 Posts	I did the linalg for 2-821 a few months back; significant oversieving meant the matrix was Sat Jun 7 13:07:34 2008 found 84366768 hash collisions in 266645965 relations Sat Jun 7 13:07:34 2008 commencing duplicate removal, pass 2 Sat Jun 7 13:44:41 2008 found 52049651 duplicates and 214596314 unique relations Sun Jun 8 17:52:47 2008 matrix is 10129794 x 10130042 (2798.1 MB) with weight 690519820 (68.17/col) and took 375 hours on a 4GB quad-core. So I'm not particularly surprised that 2^823 \pm 1 didn't take all that long ... the way msieve does filtering makes oversieving decidedly profitable.

2008-10-02, 01:28	#5
Wacky Jun 2003 The Texas Hill Country 3²·11² Posts	Bob, Greg and Bruce are, and for some time have been, MAJOR contributors to the "NFSNet effort", when they choose to be. They have also done a number of significant projects independently, or in collaboration, outside of the NFSNet framework. I think that we all have fundamentally non-conflicting goals and that a "conference" related to the independent efforts would be appropriate.