At 08:24 PM 8/30/02 -0400, you wrote: >There is a lot of interesting data in this spreadsheet. Our overall error >rate is roughly 3.5%. If you have an error-free run, the error rate is >in the 1.4% to 2% area. If you have an SUM(INPUTS) != SUM(OUTPUTS) error >or ROUNDOFF > 0.40 error that is not caused by an approaching FFT >crossover, then there is a 56% chance that your LL test will be no good!
Question, then: Assume for the sake of argument that you "watch" your results.txt files, and let's say you had THREE such errors by the time the test was 50% completed. If that 56% figure is accurate, then this hypothetical test is almost certainly no good statistically. It isn't even worth finishing the remaining 50%, is it? Would a "serious" tester try the test on different hardware, and/or would it be worth "saving" the interim files, removing them, and starting over BEFORE submitting this likely invalid result... And if the answer to the above is "yes", then why can't Prime95 be coded to simply "credit" partial time, and throw the number back in favor of a smaller exponent, a double-check, or factoring work, instead of permitting work to continue for another few weeks (or months) that is likely fruitless? I wouldn't throw back a result on just one sumout error, but on multiple ones before the test is completed.... why are we even bothering to complete it, if we KNOW the result is not only suspect, but highly so? _________________________________________________________________________ Unsubscribe & list info -- http://www.ndatech.com/mersenne/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
