Comment by villgax

And the fact that you will never be benefited from the batching they’ve explained either, you’ll be paying for every token sequential or parallel unlike nvidia or amd gpus