A Tensor Processing Unit (TPU) is a specialized hardware processor developed by Google to accelerate machine learning tasks. TPUs are custom-developed application-specific integrated circuits (ASICs) used to accelerate machine learning workloads. They are designed for a high volume of low precision computation, with more input/output operations per joule, without hardware for rasterization/texture mapping. TPUs are over 20 times faster than state-of-the-art GPUs and are specialized in deep learning tasks.
The primary task for TPUs is matrix processing, which is a combination of multiply and accumulate operations. TPUs contain thousands of multiply-accumulators that are directly connected to each other to form a large physical matrix. This is called a systolic array architecture. TPUs work by taking input data, breaking it into multiple input vectors, and then performing computations on these vectors in parallel. This allows multiple operations to be performed at the same time, which greatly increases the speed of computation.
TPUs are designed to be energy-efficient, allowing them to save power and reduce costs for their users. They are also easy to deploy and provide a complete end-to-end solution for all types of AI. To use TPUs, one needs to set up a Google Cloud account, activate the Cloud TPU API, grant Cloud TPU access to Cloud Storage buckets, and run a basic calculation on a TPU.
