Exactly right. With today's transistor budgets, the x86 ISA decoder/translator is just noise.

This is not the difference between x86 and ARM -- it's the difference between Intel's team and Apple's (also AMD's). You don't see Qualcomm being competitive even though they also use ARM.

This is not correct actually. Simpler ISA often requires bigger instruction caches, but consumes less energy because of simpler decoding logic. VLIW theoretically can be super efficient, because it discards decoding stage altogether.

In theory, yes. This is the case when you have small cores, which is why a lot of GPUs used VLIW.

But in practice the whole decoder stage is basically a rounding error because cores got so big.

What does AMD have coming in a few months?

Zen 3 on laptops. So instead of Zen 2 on 7nm, laptops should get Zen 3 on 5nm, which is both a 10% uArch clock increase, a 15+% IPC increase, and a die shrink.

Basically, laptop chips that should be around 35% faster and use less power.