On Tuesday 06 September 2005 09:23, [EMAIL PROTECTED] wrote: > > Was curious how long it would take to L-L a exponent > so large, so ran a few tests on a 3Ghz P4D. > At approx. 2.1 sec. per iteration, it would take ~41.6 > years to test the exponent M595,899,959 using a > 32,768K FFT length.
Figures not too different to the time it would have taken to L-L test a 79,000,000 range exponent using v16 on a then-state-of-the-art PIII-500. Hopefully processor speeds will have increased enough to turn ~41 years into ~4.1 weeks before exponents that size are going to be processed as a matter of course. Though I suppose someone with more avarice than common sense is going to want to start working on exponents around 332.2 million as soon as the 10M digit prize is claimed.... > That is assuming the trial-factoring to 2^80 (~1.25 > years) and the P-1 (stage 1 only where B1= 8,010,000 > B2=8,010,00 (chosen by Prime) with FFT of 32M and > 3.04% chance) of ~6 days didn't find any factors. If you had (or configured) enough memory then the B2 limit should have been much higher - around 20x B1. And the run time would be significantly bigger - around 1 year - though the chance of finding a factor during P-1 would be nearer 6% than 3%. The point here is that 32M FFT run length implies a _minimum_ working set of 256M bytes for L-L and P-1 stage 1; P-1 stage 2 needs several times that amount, 4 GBytes would seem reasonable. This _should_ be available on systems with the power to handle 500M range exponents. Assuming of course that we have all made the switch to 64-bit systems ... the address space limitations of 32 bit processors are going to be a problem long before that - indeed for some tasks like video editing the limitation of 32-bit address space is already becoming a significant problem. Personally I would have left versions of Prime95/mprime running on 32 bit processors restricted to 8M FFT size - there's still an enormous amount of work to do on exponents under 79 million. Regards Brian Beesley _______________________________________________ Prime mailing list [email protected] http://hogranch.com/mailman/listinfo/prime
