@swellman
Thank you for the PM. And I see the thread too. I'll post my tweaks (if any) today.

Thank you for your magic Max.

[QUOTE=swellman;480501]
I used these same parameters for test sieving of all polynomials over 10k Q with Q0=1e8. Yield for the above poly proved highest at 1.22. It was also the fastest, over 5% faster than the next best.
[/QUOTE]
I hope the winner doesn't change, but could you run a 1kQ test at, say, 400M to make sure the yield (or sec/rel) profile isn't dramatically different for our poly choices? Chances are decent that the winner will still be faster than any others, but it's possible that second place gets closer or even overtakes first; if so we'll have to get more detailed to see which is really faster.

If there was a clear second place, just compare the top 2 please.

@swellman
Oh well, CADO can't really optimize your poly. I list CADO output here just in case (maybe it'll test-sieve better).
Your Murphy E is record high, great job polyselecting!
Also, did you notice that c5=1092[SUP]2[/SUP]?
[code]
Y0: -46886185235594194330413354508033945923
Y1: 1130636763778943531
c0: 173790402804441583044867230320638200179414714880
c1: 8460965546400895728835520117105099940936
c2: 27918490338215558415206576404842
c3: -1257377799562289093132565
c4: -593477568335762
c5: 1192464
skew: 253519617.08179
# lognorm 62.20, E 54.15, alpha -8.05 (proj -1.82), 5 real roots
# MurphyE = 1.01282720e-14
[/code]

@Max0526 -

Thanks for attempting to optimize. The skew is a bit different. I’ll test sieve it. And no, I did not notice c5’s squareful nature.


@VBCurtis -

I will rerun the top two polys plus Max’s tweak through test sieving with 400M. The remaining polys weren’t even close.

Test Sieving Results
 

Results are below, but poly 268 (i.e. the poly from post 268) is the best performer. Max's magic seems to have helped a good bit! Apologies for the non-formatted table - it kept blowing up on me.


poly siever lpbr/a a/r Q0 alim rlim mfbr/rlambda yield (rel/Q) sec/rel # of Q sieved
268 15 33 a 100M 536M 536M 66/3.0 1.22 1.61 10000
258 15 33 a 100M 536M 536M 66/3.0 1.22 1.69 10000
251b 15 33 a 100M 536M 536M 66/3.0 1.09 1.70 10000
249a 15 33 a 100M 536M 536M 66/3.0 1.25 1.75 10000
251a 15 33 a 100M 536M 536M 66/3.0 1.14 1.81 10000
249b 15 33 a 100M 536M 536M 66/3.0 1.19 1.81 10000
247c 15 33 a 100M 536M 536M 66/3.0 1.12 1.81 10000
247b 15 33 a 100M 536M 536M 66/3.0 1.09 1.86 10000
247a 15 33 a 100M 536M 536M 66/3.0 1.15 1.89 10000
268 15 33 a 100M 400M 400M 66/3.0 1.39 2.13 1000
249a 15 33 a 100M 400M 400M 66/3.0 1.26 2.35 1000
268 15 33 a 300M 400M 400M 66/3.0 1.12 3.91 1000
249a 15 33 a 300M 400M 400M 66/3.0 1.23 4.14 1000

A much more legible table is attached. Please note that a, b, c represent the relative order of the poly within the same post. Testing was performed running 8 threads on a non-loaded and unspectacular i7.

Later sieving with 1000 Q and alim=400M are presented, and show the same relative performance. Also ran them at Q0=300M - same thing. A sieving range for Q =1k is a small range for sieving on the -a side, but there it is.

Excellent! I'll get working on the parameter selection tomorrow, after I -npr the last GPU run. I made it to almost 2.5M.

@swellman
Another one to test-sieve. As far as I remember, the smaller c5 produces the matrix that is easier to deal with.
[code]
Y0: -46886185235594194332674628035591832985
Y1: 2261273527557887062
c0: 21723798235313825239630898066777328683391747732
c1: 2115241358681729821859927675022989718073
c2: 13959248941241170772739699872412
c3: -1257377794814468498747909
c4: -1186955160520804
c5: 4769856
skew: 126970960.11264
# lognorm 61.86, E 53.81, alpha -8.05 (proj -1.82), 5 real roots
# MurphyE = 1.02050678e-14
[/code]

Same poly output by msieve:
[code]
R0: -46886185235604068970455897139592753561
R1: 2261273527557887062
A0: 12857321127755303492857855972203525455464592532
A1: 1921797384923246111297916268701787652665
A2: 30291803202582545739170753089692
A3: -1235735202657700725674501
A4: -1291101341190244
A5: 4769856
skew 126801017.94, size 4.345e-019, alpha -8.048, combined = 1.021e-014 rroots = 5
[/code]

I just ran some further test sieving on Max’s two polys posted last night.

Poly 272 seems best to date - 10K Q with Q0=100M yielded 1.25 at a slightly faster speed than poly 268. Of course that’s using the same parameters I used in post 270.

If not too burdensome, it may prove fruitful to investigate alternate parameters on poly 272.

I haven't started yet, so I'll use post 272 for parameter selection.
My teaching schedule is presently ~45 hrs from Monday 8am to Thur 4pm, so Fridays and Saturdays get most such manual labor. I went racing this past weekend, so math distractions were tabled until next weekend.