Also called an "intelligent processing unit" (IPU) and "neural engine," a neural processing unit (NPU) is a set of circuits or a separate chip that accelerates the training and execution (inference) of AI applications. See
AI training vs. inference.
The NPU performs matrix multiplication especially in desktop devices with limited computing power compared to the GPUs in datacenter servers. See
GPU,
AI PC and
AI datacenter.
In 2016, Qualcomm introduced its Zeroth Machine Intelligence Platform and Snapdragon Neural Processing Engine that enabled mobile devices to execute AI applications on their own rather than going to a datacenter in the cloud. See
neural network and
neuromorphic computing.