The instruction level parallelism in the 2-ahead branch prediction is super cool stuff, I know zen5 had this feature but it's kind of crazy that memory is so slow relative to the cores that this complexity is beneficial and worthwhile.