Polynomial Request Thread - Page 212

swellman · 2022-11-23, 22:23

I’ve asked Ed to see if he can improve Gimarel’s polynomial, primarily the very high skew. I sieved at a couple of Q values and tbh the results were a bit disappointing presumably due to the skew. But maybe Ed can improve things.

Regardless, I’ll test sieve Gimarel’s poly once Ed finishes and reports the results.

Regarding the apparent pileup on 16e_small, we got we asked for: a locally managed big siever. Now we just have to manage it.

Personally I don’t care if it takes 6 months or more to get results on “my jobs” but others may feel differently. I too have some upcoming jobs for 16e_small but that’s a discussion for another thread.

wombatman · 2022-11-23, 22:59

Quote:

Originally Posted by swellman

I’ve asked Ed to see if he can improve Gimarel’s polynomial, primarily the very high skew. I sieved at a couple of Q values and tbh the results were a bit disappointing presumably due to the skew. But maybe Ed can improve things.

Regardless, I’ll test sieve Gimarel’s poly once Ed finishes and reports the results.

Regarding the apparent pileup on 16e_small, we got we asked for: a locally managed big siever. Now we just have to manage it.

Personally I don’t care if it takes 6 months or more to get results on “my jobs” but others may feel differently. I too have some upcoming jobs for 16e_small but that’s a discussion for another thread.

I'd just be happy to get mine on one of the BOINC servers at some point. Not a huge rush, especially since it's been like 3-4 years already

chalsall · 2022-11-23, 23:04

Quote:

Originally Posted by wombatman

I'd just be happy to get mine on one of the BOINC servers at some point. Not a huge rush, especially since it's been like 3-4 years already

The fundamental currency is time.

EdH · 2022-11-24, 01:31

I PM'd this, but I'll post it here, too:

Code:

Y0: -14957978187258655870240901459296034093884
Y1: 42565897073891884679269
c0: 37746391217112409300976615136297394616548855491015
c1: 788002616915912609544957767030183328658919
c2: 7188372473469131426824096732089934
c3: -18309757962824193913492816
c4: -31525701669955896
c5: 10810800
skew: 402453809.304
# lognorm 65.65, E 57.26, alpha -8.39 (proj -3.15), 3 real roots
# MurphyE(Bf=1.000e+07,Bg=5.000e+06,area=1.000e+16)=1.551e-15

Best poly cownoise values: 562404130.06456      1.56250988e-15

The skew is only slightly less than the original and cownoise jacks it up quite a bit.

swellman · 2022-11-24, 02:43

Quote:

Originally Posted by EdH

I PM'd this, but I'll post it here, too:

Code:

Y0: -14957978187258655870240901459296034093884
Y1: 42565897073891884679269
c0: 37746391217112409300976615136297394616548855491015
c1: 788002616915912609544957767030183328658919
c2: 7188372473469131426824096732089934
c3: -18309757962824193913492816
c4: -31525701669955896
c5: 10810800
skew: 402453809.304
# lognorm 65.65, E 57.26, alpha -8.39 (proj -3.15), 3 real roots
# MurphyE(Bf=1.000e+07,Bg=5.000e+06,area=1.000e+16)=1.551e-15

Best poly cownoise values: 562404130.06456      1.56250988e-15

The skew is only slightly less than the original and cownoise jacks it up quite a bit.

Thanks Ed. I’ll test sieve both Gimarel’s original and this spun version. The spun poly with the lower skew by a good amount (and yes thereby depressing the resulting escore) may boost yield if history is any guide.

swellman · 2022-11-26, 02:40

I test sieved both Gimarel's record polynomial (post #2317) as well as EdH's spun up version (post #2325 with skew of 402453809.304) with a slightly lower e-score.

They are basically identical in terms of yield and speed. Both are superior to any and all previous polynomials, be they deg 5 or 6.

Code:

          deg 5 - Gimarel                       deg 5 - spun 
MQ          Norm_yield      speed (sec/rel)      Norm_yield      speed (sec/rel)
65             4431              0.698               3960            0.710
100            3597              0.765               3598            0.781
150            3509              0.768               3509            0.758
200            3189              0.827               3195            0.827
250            3071              0.843               3073            0.850
300            2986              0.855               2986            0.865
350            2859              0.886               2859            0.901
400            2854              0.877               2849            0.884
500            2565              0.964               2570            0.964
600            2414              1.009               2423            1.009
700            2272              1.064               2298            1.070

spec_Q range  65-685M                               65-685M

I say go with Gimarel's original poly.

BTW, I updated the record degree 5 poly for a c208 with Gimarel's recent find.

VBCurtis · 2022-11-26, 02:56

Looking at post #2315 for test-sieve params, I think you'll find better results with the smaller lim on the 3LP side: switch the 190M and 260M. If you like, I can explore this myself.

Is there interest in team-sieving e.g. 60-100M on CADO, to maximize the benefit CADO provides sieving below lim? Doing so would likely save 60-70MQ off the f-small queue, like 100-600M might be enough. I can offer my trusty old dual-10-core ivy bridge xeon as host; it can also do the matrix if Greg doesn't have available GPU time.

EdH · 2022-11-26, 03:18

I can offer some clients for a team sieve and I wouldn't mind trying to run the matrix with my GPU to see if I could, if frmky would prefer not to. My GPU is probably a bit slow, but it does have 24GB and it's faster than my 40 thread Xeons.

swellman · 2022-11-26, 11:19

Quote:

Originally Posted by VBCurtis

Looking at post #2315 for test-sieve params, I think you'll find better results with the smaller lim on the 3LP side: switch the 190M and 260M. If you like, I can explore this myself.

Please take it. Anything that can squeeze a bit more yield out of the polynomial is welcome.

Quote:

Is there interest in team-sieving e.g. 60-100M on CADO, to maximize the benefit CADO provides sieving below lim? Doing so would likely save 60-70MQ off the f-small queue, like 100-600M might be enough. I can offer my trusty old dual-10-core ivy bridge xeon as host; it can also do the matrix if Greg doesn't have available GPU time.

Yes. I will point cores at this effort.

VBCurtis · 2022-11-26, 20:54

Pretty sure the matrix won't fit in 24GB; for reference, the C207 we team-sieved in 2019 had a 72M matrix. This is why Greg should have first crack at it!

The three of us are enough to commit to 40MQ, and a fourth will likely chip in anyway. Something like 3-4 weeks should do it.
I'll do some testing this week, 34/34 against 34/35 as well as reversing the lims and a little mfb testing.

swellman · 2022-11-26, 21:45

Quote:

Originally Posted by VBCurtis

Pretty sure the matrix won't fit in 24GB; for reference, the C207 we team-sieved in 2019 had a 72M matrix. This is why Greg should have first crack at it!

The three of us are enough to commit to 40MQ, and a fourth will likely chip in anyway. Something like 3-4 weeks should do it.
I'll do some testing this week, 34/34 against 34/35 as well as reversing the lims and a little mfb testing.

Sounds good. Just let us know when and where.

Confess I may need a bit of hand holding on getting linked into someone’s machine but I’ll see if I can dig up my notes.

2022-11-26, 02:40	#2327
swellman Jun 2012 2×5×17×23 Posts	I test sieved both Gimarel's record polynomial (post #2317) as well as EdH's spun up version (post #2325 with skew of 402453809.304) with a slightly lower e-score. They are basically identical in terms of yield and speed. Both are superior to any and all previous polynomials, be they deg 5 or 6. Code: deg 5 - Gimarel deg 5 - spun MQ Norm_yield speed (sec/rel) Norm_yield speed (sec/rel) 65 4431 0.698 3960 0.710 100 3597 0.765 3598 0.781 150 3509 0.768 3509 0.758 200 3189 0.827 3195 0.827 250 3071 0.843 3073 0.850 300 2986 0.855 2986 0.865 350 2859 0.886 2859 0.901 400 2854 0.877 2849 0.884 500 2565 0.964 2570 0.964 600 2414 1.009 2423 1.009 700 2272 1.064 2298 1.070 spec_Q range 65-685M 65-685M I say go with Gimarel's original poly. BTW, I updated the record degree 5 poly for a c208 with Gimarel's recent find.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
GIMPS wiki account request thread	ixfd64	mersennewiki	169	2018-09-21 05:43
Polynomial Discriminant is n^k for an n-1 degree polynomial	carpetpool	Miscellaneous Math	14	2017-02-18 19:46
Lost Prime Raider password request thread	cheesehead	Forum Feedback	6	2009-07-28 13:02
Polynomial	R.D. Silverman	NFSNET Discussion	13	2005-09-16 20:07
Deutscher Thread (german thread)	TauCeti	NFSNET Discussion	0	2003-12-11 22:12

2022-11-23, 22:23	#2322
swellman Jun 2012 2·5·17·23 Posts	I’ve asked Ed to see if he can improve Gimarel’s polynomial, primarily the very high skew. I sieved at a couple of Q values and tbh the results were a bit disappointing presumably due to the skew. But maybe Ed can improve things. Regardless, I’ll test sieve Gimarel’s poly once Ed finishes and reports the results. Regarding the apparent pileup on 16e_small, we got we asked for: a locally managed big siever. Now we just have to manage it. Personally I don’t care if it takes 6 months or more to get results on “my jobs” but others may feel differently. I too have some upcoming jobs for 16e_small but that’s a discussion for another thread.

2022-11-26, 02:56	#2328
VBCurtis "Curtis" Feb 2005 Riverside, CA 2²×3³×53 Posts	Looking at post #2315 for test-sieve params, I think you'll find better results with the smaller lim on the 3LP side: switch the 190M and 260M. If you like, I can explore this myself. Is there interest in team-sieving e.g. 60-100M on CADO, to maximize the benefit CADO provides sieving below lim? Doing so would likely save 60-70MQ off the f-small queue, like 100-600M might be enough. I can offer my trusty old dual-10-core ivy bridge xeon as host; it can also do the matrix if Greg doesn't have available GPU time.

2022-11-26, 03:18	#2329
EdH "Ed Hall" Dec 2009 Adirondack Mtns 1528₁₆ Posts	I can offer some clients for a team sieve and I wouldn't mind trying to run the matrix with my GPU to see if I could, if frmky would prefer not to. My GPU is probably a bit slow, but it does have 24GB and it's faster than my 40 thread Xeons.

2022-11-26, 20:54	#2331
VBCurtis "Curtis" Feb 2005 Riverside, CA 2²×3³×53 Posts	Pretty sure the matrix won't fit in 24GB; for reference, the C207 we team-sieved in 2019 had a 72M matrix. This is why Greg should have first crack at it! The three of us are enough to commit to 40MQ, and a fourth will likely chip in anyway. Something like 3-4 weeks should do it. I'll do some testing this week, 34/34 against 34/35 as well as reversing the lims and a little mfb testing.