BOINC NFS sieving - NFS@Home - Page 92

swellman · 2013-11-10, 04:08

Quote:

Originally Posted by wombatman

ETA on the C168 is ~108 hours. And that's with the target density of the matrix at 100!

Impressive performance! How many threads are you running? Is this on an i7?

wombatman · 2013-11-10, 04:10

8 threads on a Core i7, yes. It uses a little over 5GB of memory!

fivemack · 2013-11-10, 12:33

Quote:

Originally Posted by wombatman

ETA on the C168 is ~108 hours. And that's with the target density of the matrix at 100!

I'm not quite sure whether you're complaining that this is a long time or impressed that this is a short time, because increasing target density usually makes things faster (if you had enough relations to get a matrix built at the higher density).

An NFS@home (so well-oversieved) C178 was 95 hours running msieve-1.52 (SVN945) under 64-bit Linux on my stock-speed i7/4770 with target density 70:

Code:

Sun Sep 22 22:45:16 2013  weight of 13120698 cycles is about 918487032 (70.00/cycle)
Sun Sep 22 22:59:40 2013  matrix is 13120526 x 13120698 (3944.6 MB) with weight 1222719217 (93.19/col)
Sun Sep 22 23:03:39 2013  linear algebra at 0.0%, ETA 103h12m

I still had the relations, so refiltering with target density 100, running on same hardware:

Code:

Sun Nov 10 15:44:16 2013  weight of 11580698 cycles is about 1158315404 (100.02/cycle)
Sun Nov 10 17:27:09 2013  matrix is 11580525 x 11580698 (4662.0 MB) with weight 1424440785 (123.00/col)
Sun Nov 10 17:31:19 2013  linear algebra at 0.0%, ETA 84h58m

A bit more memory, but a lot less runtime

swellman · 2013-11-10, 14:10

Several of us are finding 31 bit NFS@Home GNFS factoring jobs of this size take 180-200 hours to complete LA. Wombatman has compiled a version of Msieve that seems to be considerably faster.

Admittedly not an extensive study but a promising start.

I plan to run some 30 bit SNFS NFS@Home jobs using wombatman's executable. These sized jobs have been taking 25-30 hours to complete LA in the past. Hoping to see a substantial improvement.

wombatman · 2013-11-10, 17:10

I was mostly expressing surprise at the large jump from 30-bit to 31-bit. The 30-bit one I ran before was completed in ~1 day. I'll be interested to see how much of a difference the target density makes on your system. Thanks for checking!

Quote:

Originally Posted by fivemack

I'm not quite sure whether you're complaining that this is a long time or impressed that this is a short time, because increasing target density usually makes things faster (if you had enough relations to get a matrix built at the higher density).

An NFS@home (so well-oversieved) C178 was 95 hours running msieve-1.52 (SVN945) under 64-bit Linux on my stock-speed i7/4770 with target density 70:

Code:

Sun Sep 22 22:45:16 2013  weight of 13120698 cycles is about 918487032 (70.00/cycle)
Sun Sep 22 22:59:40 2013  matrix is 13120526 x 13120698 (3944.6 MB) with weight 1222719217 (93.19/col)

I've still got the relations so am trying refiltering at target_density=100, and will update this post in probably six hours when the filtering is done.

swellman · 2013-11-11, 12:27

If it is still unclaimed.

GC_5_318 is currently in LA, should finish tonight.

Thanks.

VictordeHolland · 2013-11-11, 21:09

C168_127_110 is almost ready, now in square root fase.
Taking GW_5_318 next.

wombatman · 2013-11-11, 21:14

Just an update--C168_130_119 is 50% done on the linear algebra. It would be going faster, but I use my laptop when I'm at work, so it doesn't get to churn through as much...

Still, only a few days to go.

VictordeHolland · 2013-11-11, 22:14

C168_127_110

Code:

prp65 factor: 37792525645009215672640814114905288002695134234191638876505642649
prp104 factor: 13726515009404365290893989999446372257366451228738644541888215061872548011642399835314991189895341089667

swellman · 2013-11-12, 11:05

Code:

prp90 factor: 184763462141961298781835300446977817122734239690607168398305798472772898852528335132566597
prp123 factor: 808169483508868374406392455841213056787193069554340217585858124624893209801050341692036365491419721857836553682367085563193

swellman · 2013-11-13, 02:41

Quote:

Originally Posted by swellman

I plan to run some 30 bit SNFS NFS@Home jobs using wombatman's executable. These sized jobs have been taking 25-30 hours to complete LA in the past. Hoping to see a substantial improvement.

Data point: LA of GW_4_369 had an ETA of 15 hours with 8 threads on my i7. It will finish in a few hours.

2013-11-11, 12:27	#1007
swellman Jun 2012 11×281 Posts	Reserving GW_4_369 If it is still unclaimed. GC_5_318 is currently in LA, should finish tonight. Thanks.

2013-11-11, 21:09	#1008
VictordeHolland "Victor de Hollander" Aug 2011 the Netherlands 1176₁₀ Posts	Reserving GW_5_318 C168_127_110 is almost ready, now in square root fase. Taking GW_5_318 next.

2013-11-11, 22:14	#1010
VictordeHolland "Victor de Hollander" Aug 2011 the Netherlands 2³·3·7² Posts	C168_127_110 factors C168_127_110 Code: prp65 factor: 37792525645009215672640814114905288002695134234191638876505642649 prp104 factor: 13726515009404365290893989999446372257366451228738644541888215061872548011642399835314991189895341089667

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

2013-11-10, 04:10	#1003
wombatman I moo ablest echo power! May 2013 13×137 Posts	8 threads on a Core i7, yes. It uses a little over 5GB of memory!

2013-11-10, 14:10	#1005
swellman Jun 2012 3091₁₀ Posts	Several of us are finding 31 bit NFS@Home GNFS factoring jobs of this size take 180-200 hours to complete LA. Wombatman has compiled a version of Msieve that seems to be considerably faster. Admittedly not an extensive study but a promising start. I plan to run some 30 bit SNFS NFS@Home jobs using wombatman's executable. These sized jobs have been taking 25-30 hours to complete LA in the past. Hoping to see a substantial improvement.

2013-11-11, 21:14	#1009
wombatman I moo ablest echo power! May 2013 3365₈ Posts	Just an update--C168_130_119 is 50% done on the linear algebra. It would be going faster, but I use my laptop when I'm at work, so it doesn't get to churn through as much... Still, only a few days to go.