Low precision matrix multiplication for efficient deep learning in NVIDIA Carmel processors | Publicación