no matter how quick this additional test will be, it will slow down all machines globally,
if we consider that this additional test takes only 3 minutes per exponent overall in average, ...
The test will take at most 300 clock cycles (the time to read a double from
main memory). 20,000,000 iterations * 300 clocks / 1 GHz cpu speed will add only 6 seconds to the average LL test. This is 30 times better than your 3
minutes estimate.
I think adding 6 seconds or even 3 minutes per LL test will have negligible impact on GIMPS throughput. Your analysis is correct, but maybe there are other small, unquantifiable factors at work. How about the possible cost of a second false alarm causing 3 people to drop out of GIMPS saying "these guys are bozos that don't learn from their mistakes"?
_________________________________________________________________________ Unsubscribe & list info -- http://www.ndatech.com/mersenne/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
