I don't know if you've seen Somerandombot007's thread http://www.hearthranger....sh-AI-with-procexp.aspx but the mention of 12 cores (pretty unusual) leads me to believe you have. They posted a cpu usage graph and settings today which are the highest settings I've seen anyone use and it's pegging the CPU's so it's pretty clear that silverfish can make good use out of extra cores. However, for most everything else that a typical gamer does, faster cores will provide a better experience.

I'm also starting to doubt whether higher settings = better plays at all. The simulations for the CURRENT turn are pretty good but it makes a lot of assumptions about what the opponent will do on their turn (mainly that it will always play a X/X minion where 6 >= X = mana) which causes it to make plays this turn that will trade well with the new X/X minion next turn. That can cause it to get really screwed if it was expecting a 5/5 and then a spell or something like a 4/6 is played because then it can't trade as efficiently as it wanted to. Silverfish has a built in debugging feature where you can copy a turn out of the log file, paste it into a test.txt, run silverfish.exe to see what it will do, and rerun it with different settings too. This allows us to compare certain plays with different settings at any time but it is very time consuming to go through several turns and try to decide which settings really gave the best play.

I'm just saying that we should start testing with actual boards to tell what the best settings are instead of changing them around and hoping that it helps. Right now there seems to be an assumption that higher settings will give better results but there isn't any proof. It would also be good if silverfish kept track of cards that it sees enemies play so that it could know what cards are in the meta and then simulate them as more likely to happen instead of just assuming that a X/X creature will be played.