128 TOPS, sparse acceleration.
High-efficiency NPU with 128 TOPS for edge and cloud AI inference.
64-bit out-of-order, 3.8GHz, vector extensions.