Linear algebra reservations, progress and results - Page 183

VBCurtis · 2017-07-05, 16:11

Quote:

Originally Posted by swellman

Any progress on 148^83+83^148?

remdups left 703M unique relations; I started filtering last night. I'll check on it tonight and report on a matrix size (or that density 128 failed).

swellman · 2017-07-05, 16:32

Quote:

Originally Posted by VBCurtis

remdups left 703M unique relations; I started filtering last night. I'll check on it tonight and report on a matrix size (or that density 128 failed).

Good luck! Might I ask what command you used for remdups with this job? I haven't yet run remdups on my 32 Gb machine and am not sure of the optimal parameters.

VBCurtis · 2017-07-05, 19:35

I chose 4000 for DIM, and I think the stats mentioned peak dimension used was 2500 or 2600. So, anything 3000+ should work fine; I'm new to remdups, only used it a half-dozen times or so.

fivemack · 2017-07-05, 23:16

Code:

Wed Jul  5 20:42:24 2017  p79 factor: 2353306656022930028910635861744663284129653692693621106551161170628133019327781
Wed Jul  5 20:42:24 2017  p128 factor: 35158938100505702216835191053339927789653957737747925804813852939832755829649106582970407484313288730036287017127163544287828533

194.7 hours for 17.80M matrix at density 138 (density 140 didn't work) on 6 cores i7/5820K

Log attached and at https://pastebin.com/f0isNqny

VBCurtis · 2017-07-06, 05:27

Quote:

Originally Posted by VBCurtis

remdups left 703M unique relations; I started filtering last night. I'll check on it tonight and report on a matrix size (or that density 128 failed).

Cough, the matrix built at density 128.
Cough, it's 50.0M.
Cough, 9100 hr ETA.

Trying at density 132. Do we care if this takes an actual year?

pinhodecarlos · 2017-07-06, 05:40

Quote:

Originally Posted by VBCurtis

Cough, the matrix built at density 128.
Cough, it's 50.0M.
Cough, 9100 hr ETA.

Trying at density 132. Do we care if this takes an actual year?

Better ask Greg to run it on the cluster.

fivemack · 2017-07-06, 09:45

Code:

Thu Jul  6 04:09:13 2017  p82 factor: 9736267685286991604296590214686121784068532407049693617052452318676898365463820219
Thu Jul  6 04:09:13 2017  p126 factor: 994443955602334155737530937242373089041045893579192732226709889022286276263920917709561671880531487612825000432373513246463893

About 117 hours on 7 threads E5-2650v2 for 15.06M matrix at density 130 (density 140 didn't work). Log attached and at https://pastebin.com/zzGHtnuc

swellman · 2017-07-07, 21:19

Quote:

Originally Posted by VBCurtis

Cough, the matrix built at density 128.
Cough, it's 50.0M.
Cough, 9100 hr ETA.

Trying at density 132. Do we care if this takes an actual year?

Any luck with higher TD? What are your feelings on a job duration of this magnitude?

I'm willing to try it if you decide to pass, but I doubt my hardware is any faster than yours (the opposite I predict).

VBCurtis · 2017-07-07, 21:45

Quote:

Originally Posted by swellman

Any luck with higher TD? What are your feelings on a job duration of this magnitude?

I'm willing to try it if you decide to pass, but I doubt my hardware is any faster than yours (the opposite I predict).

I filled the disk with a retry; I had moved the matrix files for TD 128 before trying 132, didn't notice they were 30+GB. I tried again last night at 136 with sufficient disk space, should have info tonight. The log indicated ~4M excess cycles at 128, which is why I thought 136 might work.

I think we can get creative about transferring the matrix files, and each run this for 5-6 months to get it complete. E.g. I'll run it until December, we figure out a way to transfer the files, and you complete the job? I think we should ask Greg to run the next GNFS, the one that's 1 digit bigger.

swellman · 2017-07-07, 22:17

I'm game, though I suspect we may encounter some issues in this process (e.g. matching superblock size, etc). Hoping others here can offer advice. Maybe a higher TD will have a significant positive effect on ETA.

FWIW, I'm trying to procure a reburbed HP Z820 with enough memory and at least partial processor power to run a 33-bit job (which will still take months to run). But I don't have it yet - waiting on funding and domestic management buy-in.

VBCurtis · 2017-07-08, 02:06

Oh, didn't think of superblock/architecture details. The CPUs are core2-era, which likely explains the leisurely ETA (dual 4-core Xeons @2.13, I think). I don't have any other machines with more than 16GB, alas; I'd be more confident if I could start the LA on the same architecture you'd be finishing it on.

I picked up a Z600 (dual 6-core Xeon) from ebay this winter, am very very happy with the purchase. 600 series has 6 DDR3 slots, 3 per CPU socket; 800 series has twice that. I do wish I'd had a crystal ball to warn me DDR prices would double 3 months after purchase, so I could have loaded it better than 6x2GB sticks. So, it doesn't do NFS@home work. A Z820 with a pile of 4GB sticks is a very capable machine!

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

2017-07-05, 19:35	#2005
VBCurtis "Curtis" Feb 2005 Riverside, CA 1001100001010₂ Posts	I chose 4000 for DIM, and I think the stats mentioned peak dimension used was 2500 or 2600. So, anything 3000+ should work fine; I'm new to remdups, only used it a half-dozen times or so.

2017-07-07, 22:17	#2012
swellman Jun 2012 2²·773 Posts	I'm game, though I suspect we may encounter some issues in this process (e.g. matching superblock size, etc). Hoping others here can offer advice. Maybe a higher TD will have a significant positive effect on ETA. FWIW, I'm trying to procure a reburbed HP Z820 with enough memory and at least partial processor power to run a 33-bit job (which will still take months to run). But I don't have it yet - waiting on funding and domestic management buy-in.

2017-07-08, 02:06	#2013
VBCurtis "Curtis" Feb 2005 Riverside, CA 2·2,437 Posts	Oh, didn't think of superblock/architecture details. The CPUs are core2-era, which likely explains the leisurely ETA (dual 4-core Xeons @2.13, I think). I don't have any other machines with more than 16GB, alas; I'd be more confident if I could start the LA on the same architecture you'd be finishing it on. I picked up a Z600 (dual 6-core Xeon) from ebay this winter, am very very happy with the purchase. 600 series has 6 DDR3 slots, 3 per CPU socket; 800 series has twice that. I do wish I'd had a crystal ball to warn me DDR prices would double 3 months after purchase, so I could have loaded it better than 6x2GB sticks. So, it doesn't do NFS@home work. A Z820 with a pile of 4GB sticks is a very capable machine!