Google’s Cloud TPU Pods are now publicly available in beta
We often see ML teams develop their initial models on individual Cloud TPU devices (which are generally available) and then expand to progressively larger Cloud TPU Pod slices via both data parallelism and model parallelism to achieve greater training speed and model scale. You can learn more about the underlying architecture of TPUs in this blog post or this interactive website, and you can learn more about individual Cloud TPU devices and Cloud TPU Pod slices here. Benchmarking one of our official reference models on a public dataset on larger and larger pod slices is a great way to get a sense of Cloud TPU performance at scale.
Source: cloud.google.com