Optimizing GPU Usage for a Kubernetes Cluster