It's basically same AMD GPU and you can compile SPIR-V into its machine code optimally without any kind of made up "I'm closer than close to the hardware" secrets. Compiler optimizations are done all the time (see Valve's aco and AMD's llvm based compiler).

Vulkan is more than adequate to handle things for the hardware on the API side so I don't buy the claim that they are somehow "closer than close" to be better.