Linear algebra reservations, progress and results - Page 161

YuL · 2017-01-20, 20:28

145^72+72^145 is factored

Code:

Fri Jan 20 14:55:09 2017  p86 factor: 87703786972364019431608195397058749092287547928991667456672843153075868703107613442793
Fri Jan 20 14:55:09 2017  p100 factor: 2210021286715014614290405621775603377151959634601407815051621353068190619444766329420677291802010633

LA on a 18.9e6 matrix took ~9.2 days using 2x 6 threads on dual Xeon E5-2620 @2.0GHz (target_density=132).
Log attached and at http://pastebin.com/km9t4KkG

VBCurtis · 2017-01-20, 21:33

Quote:

Originally Posted by YuL

LA on a 18.9e6 matrix took ~9.2 days using 2x 6 threads on dual Xeon E5-2620 @2.0GHz (target_density=132).

YuL-
When you say "2x6 threads", do you mean you used MPI to keep one set of 6 processes distinct from the other set, to keep the data local to each CPU? I ask because I have a dual-Xeon 5650 (6 cores each, 12 total) on the way, and plan on tackling some big matrices with it. Can I just invoke msieve -t 12 {etc}, or should I be mucking about with MPI and/or taskset to split the job?
I suppose "can" isn't quite what I mean; can someone quantify the speedup from a more advanced use than "msieve -t 12 {etc}"?

YuL · 2017-01-21, 07:58

Quote:

Originally Posted by VBCurtis

YuL-
When you say "2x6 threads", do you mean you used MPI to keep one set of 6 processes distinct from the other set, to keep the data local to each CPU? I ask because I have a dual-Xeon 5650 (6 cores each, 12 total) on the way, and plan on tackling some big matrices with it. Can I just invoke msieve -t 12 {etc}, or should I be mucking about with MPI and/or taskset to split the job?
I suppose "can" isn't quite what I mean; can someone quantify the speedup from a more advanced use than "msieve -t 12 {etc}"?

Here is what I do: I run two processes in a 2x1 MPI grid by running

Code:

mpiexec -np 2 --map-by socket --bind-to socket %mpi_msieve% -v -t 6 ...

one process runs on CPU1 the other runs on CPU2, it is faster than running
just one msieve -t 6 or msieve -t 12 but unfortunately I can't give you the
numbers right now.

VBCurtis · 2017-01-21, 17:39

That's exactly what I was looking for, thank you!

YuL · 2017-01-21, 20:08

Quote:

Originally Posted by VBCurtis

That's exactly what I was looking for, thank you!

You're welcome, here are some numbers:

- Using 2x 6 threads (with MPI as explained in previous post):

Code:

Wed Jan 11 09:14:20 2017  linear algebra at 0.0%, ETA 234h15m
Wed Jan 11 09:14:43 2017  checkpointing every 90000 dimensions
Fri Jan 20 13:12:01 2017  lanczos halted after 298905 iterations (dim = 18901676)
Fri Jan 20 13:12:27 2017  recovered 35 nontrivial dependencies
Fri Jan 20 13:12:29 2017  BLanczosTime: 795869
Fri Jan 20 13:12:29 2017  elapsed time 221:04:31

- Using only one process with -t 6:

Code:

Sat Jan 21 17:58:15 2017  commencing Lanczos iteration (6 threads)
Sat Jan 21 17:58:15 2017  memory use: 8046.8 MB
Sat Jan 21 18:00:09 2017  linear algebra at 0.0%, ETA 376h14m

- Using only one process with -t 12:

Code:

Sat Jan 21 18:06:00 2017  commencing Lanczos iteration (12 threads)
Sat Jan 21 18:06:00 2017  memory use: 8047.0 MB
Sat Jan 21 18:07:29 2017  linear algebra at 0.0%, ETA 293h23m

RichD · 2017-01-24, 01:47

I'll take P55 (19269...)^5-1 next.

YuL · 2017-01-24, 07:42

P55(20156...)^5-1 is factored.

Code:

Tue Jan 24 05:58:26 2017  p84 factor: 940322013842066118893478552199628075999679084425590323928948290553565322827151582101
Tue Jan 24 05:58:26 2017  p121 factor: 1537670403026780958260264404077531779849152647548875287040060122513827450125721017592918041375946498047702440132660850381

LA on a 6.23e6 matrix took ~27.7 hours using 6 threads on Core i7-5820K (target_density=136).
Log attached and at http://pastebin.com/3YLXMB1X.

unconnected · 2017-01-24, 07:59

10^249+45*10^124-1 completed - 102 hours for 10.5M matrix (TD=130)

Code:

prp97 factor: 1058314134576003349770157576757398261341275637591655186457355903095338293154242250572586542471071
prp149 factor: 47570811597419943430756398431048875252430079438673794780755089544997090197565725745187114326495767160629829316527724400223086541540148348903597430663

http://pastebin.com/3quHtP2E

YuL · 2017-01-24, 13:33

Taking C196_145_47 next.

swellman · 2017-01-24, 16:29

Reserving 19218301^31 - 1 if it is available.

RichD · 2017-01-25, 14:21

Guessing about 24 hours to solve a 6.1M matrix, -t 4, TD=130. (TD-136 failed)
The machine wasn't idle at the beginning of LA.

Code:

p94 factor: 3925488933189879342183094944089294977855388227825677716724995836344555324592472843330479439711
p116 factor: 27965361357645581827621384004007663549053808145883995338373731940837851218015431830810603685286723811342793252094611

2017-01-24, 13:33	#1769
YuL Feb 2012 Paris, France 7·23 Posts	C196_145_47 (14e) Taking C196_145_47 next.

2017-01-24, 16:29	#1770
swellman Jun 2012 2²·773 Posts	19218301^31 - 1 Reserving 19218301^31 - 1 if it is available.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
restarting nfs linear algebra	cubaq	YAFU	2	2017-04-02 11:35
Linear algebra at 600%	CRGreathouse	Msieve	8	2009-08-05 07:25
Linear algebra crashes	10metreh	Msieve	3	2009-02-02 08:34
Linear algebra proof	Damian	Math	8	2007-02-12 22:25
Linear algebra in MPQS	R1zZ1	Factoring	2	2007-02-02 06:45

2017-01-21, 17:39	#1764
VBCurtis "Curtis" Feb 2005 Riverside, CA 4874₁₀ Posts	That's exactly what I was looking for, thank you!

2017-01-24, 01:47	#1766
RichD Sep 2008 Kansas 2⁶·53 Posts	I'll take P55 (19269...)^5-1 next.