-   NFS@Home (
-   -   2018 15e post-processing reservations and results (

RichD 2018-08-31 13:43

Taking 149^53+53^149 cofactor.

RichD 2018-09-02 14:08

Lucas(2040) factored
[CODE]p80 factor: 82885753524430576798716004329371858035576140929984969068564210224542252433741601
p130 factor: 4964069304274815547360227904991595184914800703291694901664412088251250189881781301537437808398887163246559306931396793735305514881[/CODE]
211M unique relations built an 8.0M matrix using TD=124.
Solve time about 43.5 hours (-t 4).

Log at: [url][/url]

VBCurtis 2018-09-03 05:06

13_2_854m1 factors
13*2^854-1 is factored:
[code]prp73 factor: 8515676503729065430676035768998625283276009907998646812585500141457896357
prp137 factor: 92714622915185387850815972500017060454996507976891494520359406011669825965298050951631454570804864862361489807135561010514964242925791523[/code]
This was a 32LP job. 409M raw relations yielded 314.2M unique, sufficient for a Density-110 filtering pass (116 failed) to build a 16.3M matrix.
10 cores of Xeon took ~120hr to solve the matrix.
Log at [url][/url]

VBCurtis 2018-09-03 05:18

137_92 factors
137^92+92^137 is factored (GNFS-184):
[code]prp82 factor: 4445751713590772982914867485532982968766172743185249943657040471458208441687831487
prp103 factor: 2235893827625416990634491805820534937108958908510941863928368865093130364714850704924241434378644997101[/code]
457M raw relations yielded 355M unique, sufficient to build a density 128 (132 failed) matrix 19M x 19M.
10 cores of Xeon (with the other 10 hyperthreads running CADO sieving) spent 280 hr solving this matrix.
Log at [url][/url]

pinhodecarlos 2018-09-03 07:52

Curtis, have you ever tried to run LA on all HT threads as well. It helps.

VBCurtis 2018-09-03 15:44

Does taskset work with a split range, e.g. 0-9, 20-29? I haven't been using HT for matrix solving on dual-socket systems, only a desktop i7.
(I suppose I should just try it next time)

VictordeHolland 2018-09-03 17:13

[QUOTE=VBCurtis;495256]Does taskset work with a split range, e.g. 0-9, 20-29? I haven't been using HT for matrix solving on dual-socket systems, only a desktop i7.
(I suppose I should just try it next time)[/QUOTE]
The few matrices I tried with a dual Xeon E5-2650 were fastest with using numacontrol to bind the process to one of the nodes (and the memory attached to that node).[code]numactl --cpunodebind=0 --membind=0 your_msieve_name -[msieveoptions][/code]But that was with a [U]non-MP[/U]I build, so that doesn't really count...

VBCurtis 2018-09-03 19:51

Thanks, Victor! I have not yet successfully compiled msieve, so my binary is also non-MPI. I'll try numactl soon and compare speed to taskset.

pinhodecarlos 2018-09-04 07:55

I’ve no experience running msieve binary under Linux but I’ve noticed an improvement under windows when I ran -8 instead of -4 threads for LA on a HT processor.

fivemack 2018-09-06 03:02

C265_4703_73 done
Wed Sep 5 09:57:39 2018 p72 factor: 461388761568547604881966672687748898723616643167855698375696753285247229
Wed Sep 5 09:57:39 2018 p193 factor: 5585008158129255202374027022414962207693376265371023914984281687450021807020392237875954423030345959107526969415825048213766273123870914350966064821586261607308881622766314625827212253189277909

426.3hrs for 28.57M matrix on six cores i7/4930K with 128-bit vectors

Log at [url][/url] because using pastebin from a phone is fiddly; could someone please pastebin it?

fivemack 2018-09-06 03:07

Taking L4185A

All times are UTC. The time now is 10:17.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.