Thread: OBD bookmarks View Single Post
 2020-08-23, 17:33 #7 kriesel     "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 10100100000102 Posts (updated version of my previous post, past the edit expiration time) As I understand it, Factor5 is not as fast as Ernst's Mfactor program on the same hardware. And mfaktc or mfakto are much faster on a reasonable gpu (or even somewhat old slow gpu) than either of them on a normal cpu. Mfaktc can go to almost 232 exponent (1.29+ gigadigits), 95 bits factor, which seems more than enough for gigadigit candidates currently. Mfakto almost 232 exponent, 92 bits factor, also seems enough. Mfactor compiles in different variants, for speed, depending on available processor type and how much multiple precision may be needed, for very large exponents such as in the middle of https://www.mersenneforum.org/showpo...04&postcount=5 (some 66 to 127 bit exponent examples) or larger; 2, 3,4, n word precision. More info on software capabilities at the attachment of https://www.mersenneforum.org/showpo...91&postcount=2 Not intending to bash anyone's software or efforts, but there's a reference thread on Factor 5 at https://www.mersenneforum.org/showth...126#post532126 which advises using mfaktx or Mfactor instead for speed & cost when practical. Mfaktc RTX 2080 Super experience: (1 of 3 instances running together on different assignments) Code: [Fri Aug 21 02:25:10 2020] UID: Kriesel/dodo-rtx2080super, no factor for M112062857 from 2^75 to 2^76 [mfaktc 0.21 barrett76_mul32_gs] [Sat Aug 22 15:13:24 2020] UID: Kriesel/dodo-rtx2080super, no factor for M3321933179 from 2^84 to 2^85 [mfaktc 0.21 barrett87_mul32_gs] Elapsed time 84 to 85 bits, 36:48:14 as 1 of 3 instances; estimated ~12.3 hours as a solo instance. A spot check of the logs for equality of GhzD/day among the 3 instances during that period shows they're within ~3%. Power was limited to125W total for summer running. Last fiddled with by kriesel on 2020-08-23 at 17:37