Interesting that basically a decade or more of branch prediction improvements is worth a few percent in naive benchmarks of good-ness.
Also pay attention in these tables to how badly AMD missed the mark with their predecessor to Zen.