We\'re having a bit of fun here at work. It all started with one of the guys setting up a Hackintosh and we were wondering whether it was faster than a Windows Box of (nearl
I bet Miller-Rabin would be faster. If you test enough contiguous numbers it becomes deterministic, but I wouldn't even bother. Once a randomized algorithm reaches the point that its failure rate is equal to the likelihood that a CPU hiccup will cause a wrong result, it just doesn't matter any more.