MLX on silicon rocks, now imagine tuned neural engines to transformer or mamba architectures. Yes it’s just matrix multiplication, but they can optimize the chi

MLX on silicon rocks, now imagine tuned neural engines to transformer or mamba architectures. Yes it’s just matrix multiplication, but they can optimize the chips for their level of precision, memory bandwidth and unified memory. Not to mention the future of ai is an electric power battle, who outperforms everyone on performance per watt?