BOINC NFS sieving - NFS@Home - Page 57

frmky · 2013-05-14, 23:19

Quote:

Originally Posted by pinhodecarlos

Can you elaborate more? NVIDIA or AMD cards?

CUDA with CUDA-aware MPI. Beyond that, it is still on the drawing board, so no details to give yet.

pinhodecarlos · 2013-05-14, 23:43

Quote:

Originally Posted by frmky

CUDA with CUDA-aware MPI. Beyond that, it is still on the drawing board, so no details to give yet.

It's going to be an expensive LA hobbie. Combining 2, 3, 4 and 6 GB Nvidia cards to run a linear algebra...11_10_233m can only be run at least on 3+4 GB Nvidia cards or in two 4GB Nvidia cards. Here it is using 7.4 GB of memory.

jasonp · 2013-05-15, 00:31

The big clusters are increasingly relying on using heavy-duty GPUs to get major performance numbers, so we need to ride that wave.

It's not going to be easy, plus my time has been unusually tight.

RichD · 2013-05-15, 02:21

Quote:

Originally Posted by pinhodecarlos

Here it is using 7.4 GB of memory.

Hint: The whole matrix may not need to fit on the card at the same time.

henryzz · 2013-05-15, 11:45

@pinhodecarlos
11_10_233m is less sieved than the others however to make the matrix 100 hours faster to solve would need many 100s of hours more sieving. It has been noted that the larger numbers that were solving quicker had more than efficient oversieving.

pinhodecarlos · 2013-05-15, 12:18

Quote:

Originally Posted by henryzz

@pinhodecarlos
11_10_233m is less sieved than the others however to make the matrix 100 hours faster to solve would need many 100s of hours more sieving. It has been noted that the larger numbers that were solving quicker had more than efficient oversieving.

Great conclusion you reached. Thank you very much.
Have you been reading all the discussion? I don't think so otherwise you wouldn't post such comment. You need to stop posting on threads you are not at all related. It's usual from your side, you think you know everything...stop that...it's annoying.

fivemack · 2013-05-15, 12:26

Who's running L1752? If it's available I'd like to take it, since L1809 won't be ready for a little while

(though it is perhaps silly for me to use my giant stomping machine on something which has been that thoroughly oversieved; if anyone else jumps in I'm willing to hand it over)

pinhodecarlos · 2013-05-15, 13:20

Quote:

Originally Posted by fivemack

Who's running L1752? If it's available I'd like to take it, since L1809 won't be ready for a little while

(though it is perhaps silly for me to use my giant stomping machine on something which has been that thoroughly oversieved; if anyone else jumps in I'm willing to hand it over)

L1752 is free and unreserved. You can start it.

fivemack · 2013-05-15, 19:31

Started

pinhodecarlos · 2013-05-16, 23:35

Due to BOINC Pentathlon I will be delaying 11_10_233m post-processing by 2-3 days. I'll be helping team SETI.USA on Asteroids@Home project. I made this call because SETI.USA always answered to my calls to help here at NFS@Home sieving. Anyone cares to join me?

Carlos

fivemack · 2013-05-18, 14:28

Code:

Thu May 16 11:52:27 2013  found 58947591 hash collisions in 277518340 relations
Thu May 16 12:42:51 2013  found 59415645 duplicates and 218914989 unique relations
Thu May 16 17:42:02 2013  commencing 2-way merge
Thu May 16 17:42:49 2013  reduce to 20406595 relation sets and 19105130 unique ideals
Thu May 16 18:02:07 2013  found 8869502 cycles, need 8779330
Thu May 16 18:02:15 2013  weight of 8779330 cycles is about 1124196830 (128.05/cycle)
Thu May 16 18:12:51 2013  RelProcTime: 28699
Thu May 16 18:13:03 2013  commencing linear algebra
Thu May 16 18:13:03 2013  initialized process (0,0) of 3 x 8 grid

Thu May 16 19:25:14 2013  matrix is 8779153 x 8779330 (4286.8 MB) with weight 1232017598 (140.33/col)
Thu May 16 19:25:14 2013  sparse part has weight 1027182717 (117.00/col)
Thu May 16 19:33:02 2013  filtering completed in 2 passes
Thu May 16 19:33:05 2013  matrix is 8778952 x 8779129 (4286.8 MB) with weight 1232009780 (140.33/col)
Thu May 16 19:33:05 2013  sparse part has weight 1027179519 (117.00/col)
Thu May 16 19:34:58 2013  commencing Lanczos iteration
Sat May 18 12:47:51 2013  BLanczosTime: 153288
Sat May 18 14:44:42 2013  sqrtTime: 7008
Sat May 18 14:44:42 2013  prp82 factor: 1545768082218508034134592168562078966195292214045307431639948553137550491510255521
Sat May 18 14:44:42 2013  prp110 factor: 85522550180651786303284375526140009982744059479985005313382947817739050617229807903408193306696266790065760047

2013-05-15, 12:26	#623
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 2³·11·73 Posts	Who's running L1752? If it's available I'd like to take it, since L1809 won't be ready for a little while (though it is perhaps silly for me to use my giant stomping machine on something which has been that thoroughly oversieved; if anyone else jumps in I'm willing to hand it over) Last fiddled with by fivemack on 2013-05-15 at 12:28

2013-05-16, 23:35	#626
pinhodecarlos "Carlos Pinho" Oct 2011 Milton Keynes, UK 1353₁₆ Posts	Due to BOINC Pentathlon I will be delaying 11_10_233m post-processing by 2-3 days. I'll be helping team SETI.USA on Asteroids@Home project. I made this call because SETI.USA always answered to my calls to help here at NFS@Home sieving. Anyone cares to join me? Carlos Last fiddled with by pinhodecarlos on 2013-05-17 at 00:08

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Boinc Statistics for NFS@Home borked ?	thomasn	NFS@Home	1	2013-10-02 15:31
BOINC NFS sieving - RSALS	debrouxl	NFS@Home	621	2012-12-14 23:44
BOINC?	masser	Sierpinski/Riesel Base 5	1	2009-02-09 01:10
BOINC?	KEP	Twin Prime Search	212	2007-04-25 10:29
BOINC	bebarce	Software	3	2005-12-15 18:35

2013-05-15, 00:31	#619
jasonp Tribal Bullet Oct 2004 3543₁₀ Posts	The big clusters are increasingly relying on using heavy-duty GPUs to get major performance numbers, so we need to ride that wave. It's not going to be easy, plus my time has been unusually tight.

2013-05-15, 11:45	#621
henryzz Just call me Henry "David" Sep 2007 Cambridge (GMT/BST) 2·3³·109 Posts	@pinhodecarlos 11_10_233m is less sieved than the others however to make the matrix 100 hours faster to solve would need many 100s of hours more sieving. It has been noted that the larger numbers that were solving quicker had more than efficient oversieving.

2013-05-15, 19:31	#625
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 2³·11·73 Posts	Started