View Single Post
Old 2017-02-01, 22:43   #7
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

3×1,579 Posts
Default

Expanding on Wombatman's info:

Poly select can be run for as long as you like; I suggest about a GPU-week as a reasonable balance (that is, if you spent only 3 days on poly search you might spend 4 extra days on the other longer steps, compared to a better poly that the extra 4 days *might* find- it's guesswork and expectations).

Once that step is done,and you can run that step on different ranges of leading coefficients on multiple machines (say, 1 day each on 7 machines with each one assigned a range on the command line), then you'd invoke factmsieve.py with the proper number of threads for your machine's hardware. For best speed, DO take advantage of hyperthreading; a quad-core HT should have 8 threads running.

Speed: I've finished a GNFS-156 on a 6-core i7 in nearly exactly 7 days. So, using the "double every 5 digits" rule of thumb, you can expect 7-8 weeks if you have 6 cores available, or 10-12 weeks if you have just a single quad-core. You can definitely split the task over two or more machines; just make each machine do its own region of q-values.

If you look at the command that factmsieve issues to the command line as it runs, you can figure out how to manually invoke lasieve4i15e on any number of machines. It takes a little practice/trial and error; I suggest you try to factor something around 130 digits and mess with command lines and editing q-values and renaming outputs to spairs.add to see how the script adapts to these things.

If you post the number here, I'll give my GPU a few hours of poly select on it, and can explain what score means for your expected sieving time.

Note the stage after sieving, when you solve the matrix, will be done on one machine only. Expect a few days for that last step (unless you have a really fast machine like a Xeon or 6-core, which might reduce that to 50-65 hr).

Good luck!
VBCurtis is online now   Reply With Quote