Good enough for me! I'm starting with coefficient = 100,000. MSieve defaulted to a degree 6 polynomial, so I'll go with that. It'll probably be tomorrow at the earliest for a good one.

Expected e-value = from 2.82e-016 to > 3.24e-016

I don't know if degree 5 or 6 will be better; the degree-6 cutoff of 220 digits was picked ages ago and may not fit reality.

Good to know. If I don't get any suitable polys after a little bit of trying, I'll swap over to degree 5 and see if that's any better.

I've put one GTX580 on 3,697+ (starting at a6=3000000)

[QUOTE=wombatman;362486]Expected e-value = from 2.82e-016 to > 3.24e-016[/QUOTE]
Here are early returns...

# norm 4.548505e-16 alpha -10.866303 e 2.367e-16 rroots 2
# norm 4.511742e-16 alpha -10.797926 e 2.341e-16 rroots 2
# norm 4.434687e-16 alpha -10.703566 e 2.317e-16 rroots 4
# norm 4.405070e-16 alpha -10.544812 e 2.311e-16 rroots 4

24 hours on GTX-580 with standard parameters got me
[code]
R0: -404443589509309644545929847991194797
R1: 14038888150453783009
A0: 106179947038273576711479351401604537523119755826432
A1: 8293040890090289858393517925928472816068400
A2: -9872666238058635610080300055191397444
A3: -343556026632192848407094659168
A4: 162207137576070941752467
A5: 903770994047538
A6: 3003660
skew 13674337.99, size 2.798e-16, alpha -9.870, combined = 1.810e-16 rroots = 4
[/code]

so I clearly need to spend a while getting better parameters. stage1_norm=3.0e26 seems to give a more reasonable number of stage-1 hits; my new cascade looks like
[code]
~/msieve-cuda -g 0 -v -np1 stage1_norm=3.0e26
/home/nfsworld/msieve-svn/trunk/msieve -v -nps stage2_norm=1.0e28
sort -g -k11 msieve.dat.ms | head -n 10 > step3/msieve.dat.ms
cd step3
/home/nfsworld/msieve-svn/trunk/msieve -v -npr min_evalue=1e-16
[/code]

Better parameters running overnight got the psychologically-pleasing
[code]
# norm 3.761949e-16 alpha -10.071844 e 2.009e-16 rroots 4
skew: 9479495.40
c0: 99155160448856870799415795253238400285841576800
c1: -3881022030402288457804149440385549257983568
c2: -6161660345605152571229386433797287206
c3: 255822434979644537389304915227
c4: 179279820980207778939391
c5: -1679515443300779
c6: 6541260
Y0: -355240808786434010606567417147182671
Y1: 22536193199657593
[/code]

which still bears no comparison with what frmky has already. But I presume he has more than one GTX-580 and has run it for well over 32 hours.

I think I've had a lot of luck once again. 12 hours on a GTX 650 Ti got me this:
[CODE]# norm 5.999799e-16 alpha -11.867376 e 2.944e-16 rroots 6
# norm 6.156620e-16 alpha -11.156836 e 2.974e-16 rroots 4
# norm 6.106316e-16 alpha -11.918741 e 2.980e-16 rroots 6
# norm 6.230593e-16 alpha -11.925385 e 3.013e-16 rroots 6
# norm 6.290728e-16 alpha -11.141984 e 3.014e-16 rroots 4
# norm 6.320329e-16 alpha -11.011402 e 3.034e-16 rroots 6
# norm 6.373551e-16 alpha -11.171202 e 3.048e-16 rroots 4
# norm 6.513357e-16 alpha -11.084442 e 3.103e-16 rroots 6
# norm 6.554055e-16 alpha -11.204668 e 3.118e-16 rroots 4
# norm 6.660732e-16 alpha -12.359899 e 3.202e-16 rroots 4
[/CODE][CODE]# norm 6.660732e-16 alpha -12.359899 e 3.202e-16 rroots 4
skew: 80361857.79
c0: 18582064947607092585588577989168767953538872757858352
c1: -2107487898679978315922808126185301164617310220
c2: 56659332463446827797869219609467983716
c3: 1900350201082451539782799397491
c4: -10806319312797313441396
c5: -194878290326004
c6: 720720
Y0: -513064528961635260008027543725604173
Y1: 17935131075306692003
[/CODE]

My earlier results were the output of about 12 hours running on two K20's. I'm using stage1_norm=3e27, tighter than the default but not as tight as yours.

Gimarel: Nice! :smile:

Gimarel hit the jackpot. What parameters did you use?

It is a nice flare, for sure, but keep in mind that the "expected" value is a complete guess. The degree 5 and 6 e value estimates were stitched by a fudge factor, and the "expected" value is a simple (and crude) extrapolation from the much smaller input numbers. In other words, there's much to be still found - keep the engines trimmed and burning, gents!
[YOUTUBE]QjHl-57_I0g[/YOUTUBE]