It's equivalent to Tensor cores from Nvidia. This means prompt processing will get a significant boost on Apple devices, making local LLMs far more viable on future Macs (as soon as M5).