Google makes GKE dynamic slicing generally available for TPUs
AIGoogle says GKE dynamic slicing for TPUs is now generally available, decoupling logical TPU slice formation from physical nodepool provisioning. The company says it cuts AI workload startup times by up to 5x and workload recovery by 4.5x, and it is supported exclusively in TPU All Capacity under reservation mode.
Source details
Google Cloud · AI & Machine Learning · cloud.google.comPublished · added here
Original title: Now generally available: GKE dynamic slicing for higher resilience, and efficiency for TPUs