GNFS targets which need more ECM

[fivemack]

4319 curves taken through step 2 at B1=3e8 so far; I managed to break the CUDA install on the machine I was running it on. Have moved to another machine (half a GTX1080 rather than a dedicated GTX970) and expect to get to 10k by end of the month; may throw some CPU cycles at it too.

[swellman]

Originally Posted by fivemack

4319 curves taken through step 2 at B1=3e8 so far; I managed to break the CUDA install on the machine I was running it on. Have moved to another machine (half a GTX1080 rather than a dedicated GTX970) and expect to get to 10k by end of the month; may throw some CPU cycles at it too.

I've thrown more resources at this effort, should reach 10k curves by 5 Dec.

[amphoria]

Originally Posted by swellman

Do you want to tackle C193_143_93? It's next on the list.

I will take C193_143_93 for 20k curves @ B1=3e8.

[fivemack]

I am sustaining ~130 curves per hour at B1=3e8 on C192_139_81, and anticipate staying at that rate until the end of the month

[swellman]

Currently I'm at 8600 curves but my rate is much slower - more like 8.5 per hour. Not sure when we hit 20k curves but soon I think.

[fivemack]

21351 curves on C192_139_81 done on my side.

Not quite sure what our target is on this number, I'm starting to move boxes back to some of my personal sieving targets but there's a moderate number of outstanding GPU curves to post-process (I guess no more than 24 hours work on 32 threads if I schedule it sensibly); should I stop the GPU after the next block?

[swellman]

Originally Posted by fivemack

21351 curves on C192_139_81 done on my side.

Not quite sure what our target is on this number, I'm starting to move boxes back to some of my personal sieving targets but there's a moderate number of outstanding GPU curves to post-process (I guess no more than 24 hours work on 32 threads if I schedule it sensibly); should I stop the GPU after the next block?

Target was 20k curves @B1=3e8, between us we've done more than 30k curves. I'd say our work is done on C192_139_81. It's ready for sieving, though you might as well finish the last block of GPU work - a factor may appear.

I too am going to focus on other work for a bit but I will return to this thread sometime in January. Thanks to all who have contributed to this effort so far.

[amphoria]

Originally Posted by amphoria

I will take C193_143_93 for 20k curves @ B1=3e8.

20k curves completed with no factor found.

I also ran poly select for a day and a half and did some test sieving. The best SNFS candidate from yafu is

# 143^93+93^143, difficulty: 283.46, anorm: 3.30e+040, rnorm: 3.47e+052
# scaled difficulty: 285.47, suggest sieving rational side
# size = 1.325e-014, alpha = 0.000, combined = 4.818e-015, rroots = 0
type: snfs
size: 283
skew: 25.4536
c6: 1
c0: 271951251
Y1: -213816177893015973495273770237807
Y0: 175222860263437786894593195184969752945814431201

The best gnfs candidate that I found was

# norm 6.847974e-019 alpha -7.435405 e 1.200e-014 rroots 5
type: gnfs
skew: 257705007.74
c0: 505517464129968374730299942116652503025540311440
c1: 4686660945223356195158140521431875839852
c2: -86048721345626257738568839232068
c3: -104320203802295351789359
c4: 1506867183996998
c5: 1144572
Y0: -17690115030842061388581865531770840917
Y1: 378465978813130231

mieve's target range for e is 1.37e-014 to 1.58e-014 so this could be improved upon.

Both snfs and gnfs are ideally 16e jobs, but gnfs could be run with 15e and 33-bit LPs or 16e and 32 bit LPs. Yields for gnfs are 4.20 for 16e/33LP, 1.86 for 15e/33LP and 1.82 for 16e/32LP, all averages for Q over 100-700M. snfs gives a lower yield for equivalent siever and LPs.

[swellman]

Nice analysis of C193_143_93. Regardless of which NFS method is used, it appears that this will be either an extremely difficult 15e job or a "not terribly difficult to run but not currently feasible to schedule" 16e candidate. I hope one of the gatekeepers weighs in.

If NFS is not a feasible option, we can park this number for now.

[fivemack]

15/33 with a yield of around 2.0 is not at all impractical; the linear algebra will take three or four weeks on a modern eight-core Xeon or a couple of days on Greg's 160-core cluster.

But I would throw considerably more compute at the GNFS polynomial select; sieving would be twenty core-years, so two core-years of polynomial select is not excessive (or at least run a good GPU for a month). I could put 128 threads on it for a week starting 16 December if that would be helpful.

[swellman]

Originally Posted by fivemack

15/33 with a yield of around 2.0 is not at all impractical; the linear algebra will take three or four weeks on a modern eight-core Xeon or a couple of days on Greg's 160-core cluster.

But I would throw considerably more compute at the GNFS polynomial select; sieving would be twenty core-years, so two core-years of polynomial select is not excessive (or at least run a good GPU for a month). I could put 128 threads on it for a week starting 16 December if that would be helpful.

128 thread-weeks sounds fantastic. Don't have a GPU rig or I would help.

Amphoria - can you run some additional poly select? I can drop a request over in the poly select thread in the msieve project as well.