Google Cloud enables 65,000-node GKE clusters for AI training
Google Cloud developers introduced support for 65,000-node Kubernetes clusters on GKE to enable the training and inference of generative AI models with 10 trillion or more parameters. Using Terraform and Cluster Loader 2, engineers demonstrated a setup that dynamically mixes training and inference workloads on the same massive cluster, achieving pod scheduling throughput of up to 500 pods per second. Open article →