Having a hard time finding a polynomial for a C138
 

I'm currently searching a polynomial for a C138 using msieve 1.52 (SVN 942)
GPU, the best thing I could get is e = 2.831e-011 whereas the expected
e is from 2.93e-011 to > 3.37e-011. Out of curiosity I ran msieve 1.51
and it says expecting poly E from 2.55e-011 to > 2.93e-011.
Any advice(s) would be appreciated.
Note: I had two instances of msieve 1.52 running on two GPUs, the default
stage 2 norm is 1.73e+019, on the various runs I've made I set
"stage2_norm=X" with X=3e+018, 7e18, 9e18.

The "expected" scores are an estimate, which at certain sizes is difficult to achieve- you have found such a size. You found a poly with a score in the range listed from v1.51- why did you say you are having a hard time, when you do actually have a poly?

If the number takes 3 days to sieve, does it matter if you've hit the target score? Would you feel better about your GPUs if you spent 36 hr finding a poly in the v1.52 window that sieves the job 4 hrs faster than the poly you have now?

[QUOTE=VBCurtis;360626]The "expected" scores are an estimate, which at certain sizes is difficult to achieve- you have found such a size.
[/QUOTE]

Didn't know about that, kinda explains it all...

[QUOTE=VBCurtis;360626]If the number takes 3 days to sieve... [/QUOTE]

It's actually worse than that, I can do sieving + LA in less than 18 hours
:blush:

[QUOTE=VBCurtis;360626]...Would you feel better about your GPUs if you spent 36 hr finding a poly in the v1.52 window that sieves the job 4 hrs faster than the poly you have now?[/QUOTE]

Of course not, but I thought it would be possible to find a poly in the v1.52 window as I've made it for bigger numbers but then again I didn't know that the expected range is sometimes unreachable.

Thank you for your help.

There is another 'dead zone' around 150 digits.

You do have to be careful that polynomial selection not take much longer than it is worth, especially for jobs where the sieving would not take long even for a bad polynomial. The library tries to limit the total time spent searching for polynomials but that's not as simple as it sounds.

The measurement of E value was changed in v1.52 to match the scores produced by the CADO tools, but the process of selecting polynomials has not changed otherwise.