What is TPU?

TPU (Tensor Processing Unit) [1] is a chip designed by Google for machine learning, specially trained for deep machine learning, which has higher performance (per watt of computing power). Because it speeds up TensorFlow, its second-generation AI system, and is much more efficient than Gpus - Google's deep neural network is powered by the TensorFlow engine. Tpus are specifically tailored for machine learning, requiring fewer transistors to perform each operation and being naturally more efficient. Tpus can provide a higher order of instructions per watt for machine learning than all commercial Gpus and FPGas, which is basically equivalent to the level of technology in seven years. Tpus are specifically developed for machine learning applications to make chips more durable with reduced computational accuracy, meaning fewer transistors are needed for each operation, more sophisticated and high-power machine learning models are used, and those models are applied quickly, so users get more correct results

Related Products