aws cc2.8xlarge,
primecoin-0.1.1-hp5 two changes:
1) prime.cpp
const unsigned int nPrimes = (uint64)nTotalPrimes * 8 / 100; // down from 10
2) main.cpp
static const unsigned int nPrimorialHashFactor = 9; // up from 7
You seem to be testing two changes at the same time. The first change will definitely affect performance while the second change shouldn't as people have already pointed out. It's likely you're only seeing the effects of the first change you did.
mikael, knowing your work I'll take your word for it, do you agree that 1) is a change for the better and will not negatively effect block finding?
The first parameter he changed definitely has a huge impact on performance, that's for sure. The parameter is based on sieve round limit implemented by Sunny King. As you probably know, that resulted in a pretty big performance boost. I'm using a static percentage value in my code instead of counting microseconds.
Changing that percentage may indeed improve performance on some system. It depends on the efficiency difference between the sieve and the primality tester. I may change the percentage into a configuration parameter in the future.