Tensor Processing Unit is an ASIC tailored specifically for the matrix multiplication operations central to neural networks. Unlike general-purpose CPUs or GPUs, TPUs strip away unnecessary control logic to pack massive arrays of multiply-accumulate units, drastically increasing throughput and power efficiency for AI inference and training.
The edge device uses a low-power TPU to handle the image recognition locally without hitting the cloud.
Also heard as
- Tensor Processing Unit
Related terms
ASIC
An application-specific integrated circuit designed for one dedicated task rather than general-purpose use.
Graphics processing unit
A specialized processor designed to rapidly manipulate memory to accelerate image creation.
Central processing unit
The primary silicon processor responsible for fetching, decoding, and executing instructions in a computer system.