Run Ray on TPU, Part 1: The foundations- Google Developers Blog

This post introduces the first-class integration of Google Cloud TPUs into the Ray distributed computing framework. It details how Google Kubernetes Engine (GKE) provisions and labels TPU slices, and how Ray Core's `slice_placement_group` function reserves these slices atomically. The post explains the concept of TPU slices and topologies, and how GKE's Ray Operator add-on, including the Ray TPU webhook, facilitates this integration. It highlights that existing Ray code for GPUs can now run on TPUs with minimal changes, focusing on the foundational aspects of GKE and Ray Core for TPU orchestration.