Hi, At 06:19 PM 5/25/2002 -0400, Nick Glover wrote: >I noticed that v22.2 and v22.3 automatically do roundoff checking every >iteration for any exponent close enough to the FFT limit. Is there any >reason to be concerned about the possibility of roundoff error for CPUs >that aren't P4s?
Yes, a tiny bit. >What about if the non-P4s are only doing double checks? >Very specifically, I'm wondering if I should be ok if I use the >"undocumented" setting in prime.ini to turn off roundoff checking every >iteration for when my Pentium 200 MHz double checks 6502049 ( the next FFT >size is at 6520000 ). This may be a long-winded answer.... First off, the latest 22.3 version has adjusted the FFT crossovers. The P4 crossovers are more conservative than v21 and the non-SSE2 crossovers are more aggressive than before. In fact, the non-SSE2 crossover is tentatively set at 6542000. With error checking auto-enabled for exponents above 6509290. So, for your exponent, you can either safely turn off the extra error-checking or give 22.3 a test-drive. The reason 22.3 does error checking every iteration is that it allows prime95 to handle roundoff errors up to 0.6 (instead of 0.5) without corrupting the results. This in turn lets us be a little more aggressive in the FFT crossovers. Even if you are assigned an exponent that v22.3 wants to roundoff check every iteration, you can still turn off the extra error checking. If we do our job correctly you are very unlikely to get a deadly round-off error greater than 0.5. The extra error checking just increases your safety margin. Brian Beesley and I are still working on the finalizing these FFT crossover points, so they are still subject to some change. However, I can definitively state that the P4 crossover points in v22.3 are much better than in v21. P4 owners will have to upgrade to version 22 at some point in time. P3 owners will not have to upgrade to v22 if they do not want to. v22.3 is still in beta and can be downloaded from ftp://mersenne.org/gimps/p95v22.zip -- George _________________________________________________________________________ Unsubscribe & list info -- http://www.ndatech.com/mersenne/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
