View Single Post
Old 2017-02-02, 17:19   #14
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

2×2,309 Posts
Default

Sorry I forgot the worktodo.ini part- glad you solved that one!

My settings do not create much output into .ms; I think four GPU-hours found ~90 lines in the .ms file, something on the order of one every 4-5 minutes.

The -npr command works on the raw data in the .ms file, outputting actual usable polynomials into RSA.p. The best-scoring poly found will be saved to msieve.fb; in order to begin another run, you'll have to rename that file (I use filenames like RSA{scoreofthispoly} so I can see from the directory which poly had the best score, so the one I found for you is saved as RSA346).

I don't know why the script wasn't using your GPU fully, but I can tell you the script predates msieve-GPU code. I also can tell you that the -npr step takes a lot of time on each hit, and the default settings in msieve generate a LOT of hits into .ms! Perhaps the script tries to do -npr on all of them, pausing the GPU as it processes all of the GPU's output. My custom norm settings reduce the output from the GPU to just the very best candidates in the raw data, so that -npr takes perhaps an hour or two per day of GPU output.

If you have a fast GPU, say 900 series or newer, you may wish to add -t 2 after the -np1 -nps flags; this sends two sets of data to the GPU to process at the same time (two threaded, so to speak). This helps with smaller input numbers, say C150s, but is likely only slightly useful at this size.

As for running from 0 to 400, the default behavior is to start with coefficient 0 and work up, and really small coeffs take a while to churn through. I often start my search at 5000 or 10000; I only selected 5 million for mine to avoid overlapping any work you had already done. Smaller coeffs have a history of producing good scores with slightly less search time than large ones; if you run 10000 to 500000 with my norm settings, I bet you'll find a few polys above 3.2e-13.

Yes, your pastebin output is exactly what should be in .ms; if you ran ./msieve -npr -s RSA msieve would process that raw polynomial into one that the script can run sieving on (a list is saved to RSA.p). Not every raw poly line will produce a good-enough poly to save into RSA.p; I'd wait for 50 or more lines in RSA.ms to bother stopping the GPU search to run -npr. Note that once you run -npr, you should rename RSA.ms into something else, so that you don't re-process those hits the next time you run -npr. Also, change the input coefficient to start where the GPU left off; if your search got from 6000000 to 6150328, I'd start the next run at 6150400. But, start at 10000 sometime.
VBCurtis is offline   Reply With Quote