They are still somewhat in the domain of performance. * ML is used for branch prediction,[1] as well as [implied] pre-fetch. * Automatic clock and power tunin

They are still somewhat in the domain of performance. * ML is used for branch prediction,[1] as well as [implied] pre-fetch. * Automatic clock and power tuning.[2] This is not per-die, but granular across the chip. * eXtended Frequency Range,[3] unlocks the ranges that [2] is allowed to use. Essentially, the chip overclocks itself when you provide it with more cooling headroom. [1]: https://youtu.be/X9NNOqzTbKI?t=13m53s https://youtu.be/X9NNOqzTbKI?t=13m53s [2]: https://youtu.be/X9NNOqzTbKI?t=14m38s https://youtu.be/X9NNOqzTbKI?t=14m38s [3]: https://youtu.be/X9NNOqzTbKI?t=15m22s https://youtu.be/X9NNOqzTbKI?t=15m22s