I need to play a bit with it but two sample runs produce nice, high detail results - even if the samples from each run are very similar
