Basically getting around the branch predictor problem with generalized compute architectures https://en.wikipedia.org/wiki/Branch_predictor