Google recently introduced new Cloud TPU Virtual Computers (VMs), which allow users to connect directly to TPU host machines. The company provides a new and better user experience for developing and deploying TensorFlow, PyTorch, and JAX on Cloud TPUs with these VMs.
Customers can already use the TPU chipsets to create virtual instances on Google Cloud. However, because the instances were not running in the same server environment, there were some limitations. Because applications had to transfer data over the network to a TPU and then wait for the processed data to be returned back, the TPUs were connected to the chipsets remotely via a network connection, limiting processing speed.
To Read More: InfoQ