It’s fairly expensive but well suited to pipelined implementations in hardware circuits: https://medium.com/@himanshu0525125/global-history-branch-pr.... Modern CPU branch predictors can deliver multiple predictions per clock cycle.