Dissecting the Apple M1 GPU, the end

(Why) are the M3 and M4 really that different from M1 and M2?

They’re significantly different GPU architectures. They added support for hardware features like mesh shading, raytracing and better shader occupancy/dynamic caching.

Beyond that, each M series generation also brings more of the system into the SoC. For example, the entire storage controller is part of the SoC in the M1, but the M2 brought in the trackpad controller as well.

Bringing more functionality into the SoC has many advantages but it does make it more difficult to target because you can’t just make use of existing off the shelf controller knowledge to apply to it.