Thanks for the links. We all understand economical effect: faster hashing will be equivalent to faster ASICs, complete crack will mean quick adaptation of another hash (e.g. SHA-3). My question was not about that, but about explorations in how to make partial collisions faster (I do not imply "instantly") for double SHA. Are there known papers/people who explore it? I'm just curious whether in some future there will be a very bitcoin-specific method to predict, say, if you are going to have some zeros in the output or not. So you save some cycles and jump to next nonce. The algorithm can be probabilistic, of course. E.g. if we can figure out some patterns in the internal state during hash computation.