mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > NFS@Home

Reply
 
Thread Tools
Old 2021-10-04, 13:30   #67
charybdis
 
charybdis's Avatar
 
Apr 2020

547 Posts
Default

Quote:
Originally Posted by swellman View Post
However the normalized results tell a different story:
The normalized relations formula is not applicable when composite special-q are being used, as it is based on the premise that special-q have the same frequency as primes.
charybdis is offline   Reply With Quote
Old 2021-10-04, 13:53   #68
swellman
 
swellman's Avatar
 
Jun 2012

2×53×13 Posts
Default

Quote:
Originally Posted by charybdis View Post
The normalized relations formula is not applicable when composite special-q are being used, as it is based on the premise that special-q have the same frequency as primes.
Thanks for this. So the actual yield when not using the -d 1 flag is very high for the quartic but then we would likely have many WUs return partial or no results. Unless Greg has a later version of ggnfs 16 which deals with the matter cleanly. Or something else entirely. [/my pointless speculation]
swellman is online now   Reply With Quote
Old 2021-10-04, 14:48   #69
charybdis
 
charybdis's Avatar
 
Apr 2020

10001000112 Posts
Default

Using composite special-q will also lead to a high duplication rate. It's a useful way to squeeze out a few more relations when yield is low due to suboptimal parameters (eg artificially low lpb and lims on big NFS@Home jobs) but should probably be avoided if possible.
charybdis is offline   Reply With Quote
Old 2021-10-04, 14:52   #70
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

3×19×89 Posts
Default

I used -d 1 because I believe that is how the f-small siever runs. I don't think you have a worry about the failed workunits, but Greg should comment before we submit the job.
VBCurtis is offline   Reply With Quote
Old 2021-10-04, 15:49   #71
frmky
 
frmky's Avatar
 
Jul 2003
So Cal

2×3×7×53 Posts
Default

The NFS@Home binaries use only prime special-q.
frmky is offline   Reply With Quote
Old 2021-10-04, 18:33   #72
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

3×19×89 Posts
Default

Quote:
Originally Posted by swellman View Post
Once I got the updated siever running, first thing was to duplicate your earlier results. Identical output. So then I repeated the test sieving using blocks of 10 kQ. Results below:

Code:
#Q=75M  34363 rels, 556 spq, 34084 norm_rels
#Q=100M 31943 rels, 551 spq, 31472 norm_rels
#Q=150M 29363 rels, 531 spq, 29373 norm_rels 
#Q=200M 28236 rels, 529 spq, 27925 norm_rels
#Q=250M 26102 rels, 510 spq, 26468 norm_rels 
#Q=300M 26396 rels, 526 spq, 25709 norm_rels
#Q=350M 25044 rels, 510 spq, 24960 norm_rels
#Q=400M 24646 rels, 511 spq, 24350 norm_rels
#Q=450M 22276 rels, 477 spq, 23438 norm_rels
Smooth yield, but as you pointed out we don’t need to tweak the Q-range for sieving as we really don’t know our target number of relations. Greg has suggested 1B to me, but I adjusted this down to 900M with my original poly. But I agree that we can likely drop this further. Upward adjustment is always a later option.
How about Q=60-375M to start? Your yield results match mine pretty well, and Greg has confirmed prime spQ so these numbers should be accurate. I can do the matrix if Sean or Greg don't want it; I like these new-ground jobs.
VBCurtis is offline   Reply With Quote
Old 2021-10-04, 18:49   #73
swellman
 
swellman's Avatar
 
Jun 2012

2×53×13 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
How about Q=60-375M to start? Your yield results match mine pretty well, and Greg has confirmed prime spQ so these numbers should be accurate. I can do the matrix if Sean or Greg don't want it; I like these new-ground jobs.
Sure - specQ range shifted to 60-375M.

And thanks for taking on the LA! Should be fun.
swellman is online now   Reply With Quote
Old 2021-10-07, 16:58   #74
jyb
 
jyb's Avatar
 
Aug 2005
Seattle, WA

2×34×11 Posts
Default C199_M31_k35

With all of the discussion about this difficult quartic and how best to sieve it, there's something I'm missing: why aren't we using the obvious sextic, which has a very reasonable difficulty of 280, instead?
jyb is offline   Reply With Quote
Old 2021-10-07, 17:04   #75
swellman
 
swellman's Avatar
 
Jun 2012

2·53·13 Posts
Default AS C202_3408_1693

QUEUED AS C202_3408_1693

I was cleaning out the attic and found AS 3408:i1693 with a record setting e-score poly. It has been sitting on a shelf for almost a year. I'll place it here for enqueing at some point after the f_small siever catches up.

Code:
n: 7610429701878258146556615560847196204170260478136688627519823621252186450051556772508489716641675238873878778627282501333960153571054969424187320132140981530532671111226675538849711076698014563985929827
# norm 1.168186e-19 alpha -8.419964 e 4.117e-15 rroots 5
skew: 45645857.17
c0: -18098601315886851468557887762291980392541100584
c1: 4841688835906037944690945208990931966764
c2: -175894955102251933301443112313837
c3: -5578183143934717988864129
c4: 61119537791822196
c5: 908107200
Y0: -631670594992711929340967533254444051034
Y1: 75282186386799330835453
rlim: 225000000
alim: 225000000
lpbr: 33
lpba: 33
mfbr: 66
mfba: 96
rlambda: 3.0
alambda: 3.7
Test sieving results on the -a side with Q in blocks of 10k:
Code:
Q0(M)   Norm_Yield
60        33339
100       33127
150       30001
200       28658
250       26478
300       25448
350       23827
This suggests sieving with specQ range of 60-360M should produce 850M relations. A good start at least.

Last fiddled with by swellman on 2021-10-28 at 01:07
swellman is online now   Reply With Quote
Old 2021-10-07, 17:12   #76
swellman
 
swellman's Avatar
 
Jun 2012

325010 Posts
Default

Quote:
Originally Posted by jyb View Post
With all of the discussion about this difficult quartic and how best to sieve it, there's something I'm missing: why aren't we using the obvious sextic, which has a very reasonable difficulty of 280, instead?
The explanation is simple: the poly generator at cownoise did not produce a sextic. It often does for various exponents but not in this case. I assumed none existed but thank you very much for seeing it.

What is the sextic? A S280 is likely far less difficult than a S261 quartic but it likely still belongs on the f_small siever.
swellman is online now   Reply With Quote
Old 2021-10-07, 17:20   #77
jyb
 
jyb's Avatar
 
Aug 2005
Seattle, WA

2×34×11 Posts
Default

Quote:
Originally Posted by swellman View Post
The explanation is simple: the poly generator at cownoise did not produce a sextic. It often does for various exponents but not in this case. I assumed none existed but thank you very much for seeing it.

What is the sextic? A S280 is likely far less difficult than a S261 quartic but it likely still belongs on the f_small siever.
Code:
n: 4950036370987531796596777020435431752821658735009564665203032637182311299258729058872170618265983042169789410985851232822128958988972946232939020745222679656170411937888556774248091592015827689480701
skew: 1.0000
c6: 1
c5: 1
c4: 1
c3: 1
c2: 1
c1: 1
c0: 1
Y1: 1
Y0: -45671926060252476630107084286792841360213803007
type: snfs
jyb is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Queue management for e_small and 15e queues VBCurtis NFS@Home 222 2021-12-06 17:29
Queue management for 14e queue VBCurtis NFS@Home 72 2021-12-02 17:10
Run down the queue on MPRIME without quitting GIMPS Rodrigo Software 7 2018-05-25 13:26
Improving the queue management. debrouxl NFS@Home 10 2018-05-06 21:05
split a prime95 queue & client installation joblack Information & Answers 1 2009-01-06 08:45

All times are UTC. The time now is 13:14.


Tue Dec 7 13:14:58 UTC 2021 up 137 days, 7:43, 0 users, load averages: 1.15, 1.18, 1.37

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

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.