- M3, M3 Pro, M3 Max
- A17 Pro
- M4
- [1]
- Dynamic shader core memory which allows an app to achieve better thread occupancy (2 virtual threads in one).
- Flexible on-chip memory - 3 cache types now is dynamically allocated, if one type is not used - it will not be allocated.
- fp32, fp16, int can execute in parallel.
- Optimized for fp16.
- Shader execution reordering.
- In mesh shading: intermediate data from task shader now stored in the on-chip memory and accessed by mesh shader.
- Hardware ray tracing.