View Single Post
Old 2016-10-18, 23:20   #8
GP2
 
GP2's Avatar
 
Sep 2003

5·11·47 Posts
Default

I would be interested in P-1 or P+1 save files for small exponents.

For exponents in the low single-digit-thousands, I think these files should be only a few hundred bytes long. The intermediate files (.bu or .bu2) might be very large (hundreds of MB), but the non-bu final savefile should be really tiny for those very small exponents, assuming the P-1 or P+1 test actually ran to completion.

Amazon EC2 cloud has some instances with large amounts of memory, all the way up to x1.16xlarge with 976 GiB and 32 cores, and x1.32xlarge with 1952 GiB and 64 cores. The spot prices are about $0.70/hour and $1.50/hour for these, and there are others with less memory for lower cost. I have no idea how long stage 2 might take with GMP-ECM, but if it's days rather than months then something might be feasible.

When you guys did your tests, were you constrained by available memory or by CPU time, or both?

Sadly, openmp does not seem to work properly with P-1 for GMP-ECM, so only one thread could be used per exponent. If this bug could be fixed then perhaps it might go faster.

Last fiddled with by GP2 on 2016-10-18 at 23:21
GP2 is offline   Reply With Quote