Tobias Mann / The Register ·
AMD acquires Toronto-based Taalas, which integrates model weights directly into silicon to boost inference performance, for an undisclosed sum
Early tech demos show model-specific integrated circuits churning out up to 17,000 tokens a second