Reserved for MF - Sequence 4788

[bsquared]

Data attached, if anyone is interested.

[henryzz]

Originally Posted by henryzz

once i have finally managed to compile msieve-gpu i will have a CUDA gpu i can run on polynomial selection
it is proving difficult to get environment variables and such like correct

all you need is a nvidia series >=8
those look good polys

i finally got it working
i will start polynomial selection with it

[axn]

Originally Posted by bsquared

So on a whim I drug out a 1.92e-12 scoring poly with a better alpha (-6.44), and it beat the pants off of the 2.14e-12 poly. <snip>
I think there are better polys to be found, but this is the baseline so far. Lesson to be learned: Murphy_E is a crapshoot!

May I offer the following three polys for your consideration:

BEGIN POLY #skewness 492944.87 norm 2.47e+021 alpha -6.03 Murphy_E 2.00e-012
X5 22783680
X4 3049889691192
X3 25115676975823063898
X2 -1057549120220075124773221
X1 -4015168516778629810185220755126
X0 -68074828915752550906211652474462780
Y1 56460946476157681
Y0 -548263756929827444068662381961
M 130374183394572597071227248002684258680753480045928203593454005872997714178038084759490414258234615905198731655011653462453390100735529354844596548096485215
END POLY
BEGIN POLY #skewness 591796.65 norm 3.01e+021 alpha -6.32 Murphy_E 2.02e-012
X5 22783680
X4 2209855409592
X3 25038106255076701466
X2 -1612254100171446541836739
X1 -7285061836793671322852863008672
X0 120437822537708031145046391210398448
Y1 56460946476157681
Y0 -548263757346170463383849121655
M 130374183394572597071227248002684258680753480045928203593454005872997714178038084759490414258234615905198731655011653462453390100735529354844596548096492589
END POLY
BEGIN POLY #skewness 594103.05 norm 3.05e+021 alpha -6.34 Murphy_E 2.03e-012
X5 22783680
X4 2558901387192
X3 25067329196727393818
X2 -1381972795712101042561075
X1 -7294236258178618545696273107894
X0 -164061303268787107813228616719980278
Y1 56460946476157681
Y0 -548263757173174123380901987071
M 130374183394572597071227248002684258680753480045928203593454005872997714178038084759490414258234615905198731655011653462453390100735529354844596548096489525
END POLY

[axn]

Originally Posted by bsquared

details:

#skewness 979863.43 norm 8.85e+20 alpha -5.12 Murphy_E 2.14e-12
c5: 156600
c4: 5511109690356
skew: 197324.88 
 
#skewness 1249264.69 norm 1.13e+21 alpha -5.19 Murphy_E 2.07e-12
c5: 156600
c4: 5314399732356
skew: 197324.88 
 
#skewness 1239920.67 norm 1.08e+21 alpha -5.17 Murphy_E 2.07e-12
c5: 156600
c4: 5335915006356
skew: 197324.88

Saaay... Did you just test all those goodly polynomials with wrong skew???

EDIT:- Yes you did For the 2.14:
Wrong skew {total yield: 1050, q=30001003 (0.33863 sec/rel)}
Proper skew {total yield: 1174, q=30001003 (0.29744 sec/rel)}

[bsquared]

Originally Posted by axn

Saaay... Did you just test all those goodly polynomials with wrong skew???

EDIT:- Yes you did For the 2.14:
Wrong skew {total yield: 1050, q=30001003 (0.33863 sec/rel)}
Proper skew {total yield: 1174, q=30001003 (0.29744 sec/rel)}

Err, yes, yes I did . Copied it over from the last C157...

I get:

Wrong skew {total yield: 1116, q=10001009 (0.17081 sec/rel)}
Proper skew {total yield: 1341, q=10001009 (0.14201 sec/rel)}

FWIW, the 1.92e-12 poly was also tested wrong:

Wrong skew {total yield: 1396, q=10001009 (0.14012 sec/rel)}
Proper skew {total yield: 1421, q=10001009 (0.13689 sec/rel)}

And here are your 3 candidates (as you posted them, top to bottom):

total yield: 1078, q=10001009 (0.13871 sec/rel)
total yield: 951, q=10001009 (0.14025 sec/rel)
total yield: 1211, q=10001009 (0.14852 sec/rel)

And my two 2.07e-12 polys:

total yield: 1195, q=10001009 (0.14133 sec/rel)
total yield: 1038, q=10001009 (0.14251 sec/rel)

And finally, jrk's msieve poly:
total yield: 1163, q=10001009 (0.14720 sec/rel)

I'll take the 2.14 poly, 1.92 poly, and your fastest poly and do some more trial sieving.

[bsquared]

Originally Posted by bsquared

I'll take the 2.14 poly, 1.92 poly, and your fastest poly and do some more trial sieving.

A winner is emerging: axn's 2.00e-12 poly appears to be consistently faster, with a higher yield, than the others I tested.

n: 1128681916333165583281832717120127496638972860118923964844791402387721331170448754199775637678274646951935918929237315749341487244542351920306209843985409563
#BEGIN POLY #skewness 492944.87 norm 2.47e+021 alpha -6.03 Murphy_E 2.00e-012 
c5: 22783680 
c4: 3049889691192 
c3: 25115676975823063898 
c2: -1057549120220075124773221 
c1: -4015168516778629810185220755126 
c0: -68074828915752550906211652474462780 
Y1: 56460946476157681 
Y0: -548263756929827444068662381961 
skew: 492944.87
rlim: 45000000
alim: 45000000 
lpbr: 29 
lpba: 29 
mfbr: 58 
mfba: 58 
rlambda: 2.6 
alambda: 2.6

[jrk]

Here are the three best polys I found overnight:

# norm 2.999613e-15 alpha -7.053170 e 2.102e-12
skew: 4947197.07
c0:  254361561800804638444888132008691713600
c1:  277567242655343801120663754660120
c2: -41169227890365677403210934
c3: -33865259805947197259
c4:  5813164911518
c5:  214800
Y0: -1393498408963548236214652730759
Y1:  581416348390623701
# norm 2.797246e-15 alpha -7.634500 e 2.004e-12
skew: 7829822.03
c0: -392472115828519587544012662072667953792
c1:  1135538850431039541173288486836512
c2:  215718619935831247136208602
c3: -66702913924475596555
c4:  4019834079518
c5:  214800
Y0: -1393499379793961455729608702127
Y1:  581416348390623701
# norm 2.691749e-15 alpha -7.287346 e 1.958e-12
skew: 7493033.63
c0: -1191940646279096538567449411970147312640
c1:  245517647645424865764025430284152
c2: -67767123827532957236661598
c3: -26904687145857425547
c4:  6126231615518
c5:  214800
Y0: -1393498239483008345741406384063
Y1:  581416348390623701

And look at the alpha on this one!

# norm 2.644805e-15 alpha -8.500494 e 1.914e-12
skew: 18761259.11
c0:  70384025424543567250040749496664182714880
c1:  34739194678189905179867695959241928
c2:  369205076515588254645800050
c3: -294264765478046013769
c4:  659280547786
c5:  214200
Y0: -1394280990497229313048382187377
Y1:  593607538600963639

[bsquared]

Originally Posted by jrk

Here are the three best polys I found overnight:

I'll do a comparison.

[10metreh]

I always hate it when no-one mentions the siever used for testing, or bothers to test between sievers. I presume you have been using 14e for testing, but it would be good to give 15e a try, just in case.

[bsquared]

The clear winner is now this one from jrk:

 
n: 1128681916333165583281832717120127496638972860118923964844791402387721331170448754199775637678274646951935918929237315749341487244542351920306209843985409563
# norm 2.797246e-15 alpha -7.634500 e 2.004e-12
skew: 7829822.03
c0: -392472115828519587544012662072667953792
c1:  1135538850431039541173288486836512
c2:  215718619935831247136208602
c3: -66702913924475596555
c4:  4019834079518
c5:  214800
Y0: -1393499379793961455729608702127
Y1:  581416348390623701
rlim: 45000000
alim: 45000000 
lpbr: 29 
lpba: 29 
mfbr: 58 
mfba: 58 
rlambda: 2.6 
alambda: 2.6

I didn't test the last one with really low alpha.

Anyone still have any poly selection going on? Henryzz?

The above poly would need ~ 39MQ to get ~50M relations, and take a little less than 6Msec. We've chopped a million seconds of sieving time off by finding better polys (and using them correctly)... pretty nice!

[edit]
Using 14e siever. sorry I thought that was clear. I don't think we need 15e for this one... Feel free to test it if you want.

[henryzz]

i stopped my poly selection run
i suspect my gpu is rather pathetic
also it seg faults every time i hibernate which makes it almost unusable for that size number