On 17 Feb 2002, at 17:54, Russel Brooks wrote: > Am I interpreting this thread correctly? That more factoring is > needed? My climb up the LL top producers is starting to stall > so maybe it's time to switch to factoring.
I'm quite sure there's no need to panic! So far as I'm concerned, I've been underdoing factoring - less than 5% of my contribution. Theoretically the best balance for the project as a whole is about 10% of the CPU effort in trial factoring. The other point here is that we have had two major improvements in the efficiency of the LL testing algorithm since the last time trial factoring was seriously looked at, so some theoretical work on trial factoring is probably due. Why not try some factoring assignments, if _you_ think it will be fun for a change? You can always change back when you get bored... > > I know the LL tests drive the FPU hard, what about factoring? Factoring is a light PFU and memory load. LL (and DC) assignments are heavy on both. Indications to run factoring irrespective of CPU speed: 1) One CPU on a dual processor system. Both processors running LL tests hits the memory bus too hard & ruins performance. 2) Some processors (AMD K6, and especially Cyrix 6x6) have inefficient FPUs; it is much more effective to have these processors running trial factoring. 3) A PIII or Athlon system running very close to the limit of the cooling system. Because the load on the FPU imposed by factoring is much lower, the power consumption of the CPU will reduce if factoring is run instead of LL testing, and the temperature will therefore drop significantly. 4) System very short of main memory. Trial factoring uses very little memory for data work space. Counterindications: It doesn't seem to be sensible to be running trial factoring on a P4. The P4 architecture is _very_ efficient at running LL tests, but the performance on the instruction mix used in trial factoring is a great deal worse (clock for clock) than PII/PIII/Athlon. Regards Brian Beesley _________________________________________________________________________ Unsubscribe & list info -- http://www.ndatech.com/mersenne/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
