@m20k #inorder+#prefetch is fantastic for energy efficiency, but #OoO’s flexibility in the face of cache misses and #branchmissprediction leads to much better performance for many workloads. By “scaling” I meant scaling to higher performance with wider issue, requiring even better BP accuracy and covering for even longer #cachemiss penalties.

We can do better than scaling this instruction-at-a-time model we have been using since the 1950es, but that’s for another toot.