Outside of prompt processing, the only reason GPU's are better than CPU's for inference is memory bandwidth, the performance of apple M* devices at inference i

Outside of prompt processing, the only reason GPU's are better than CPU's for inference is memory bandwidth, the performance of apple M* devices at inference is a consequence of this, not of their UMA.