Feature description
Has the current Metal 4 already been implemented to support the new Tensor APIs for Neural Accelerators on Apple Silicon M5?
Feature motivation
https://www.macrumors.com/guide/m4-vs-m5-chip/
- 4×+ peak GPU compute performance for AI
- 3.6× faster time to first token (LLM)
(Optional) Suggest a Solution
https://developer.apple.com/metal/whats-new/
Metal 4 machine learning
Tensors are now a native resource type you can use to operate on data.