50. Accelerators and AI workloads

Schedule GPUs and other accelerators through device plugins, node labels, quotas, and topology-aware placement. Design training and inference workloads around large images, shared storage, batch queues, model serving, and expensive capacity.