Kubernetes documentation - Schedule GPUs

Fetched 2026-10-04 (page footer: last modified September 20, 2024). Rules verbatim: 'GPUs are only supposed to be specified in the limits section, which means: You can specify GPU limits without specifying requests, because Kubernetes will use the limit as the request value by default. You can specify GPU in both limits and requests but these two values must be equal. You cannot specify GPU requests without specifying limits.' Heterogeneous fleets: 'If different nodes in your cluster have different types of GPUs, then you can use Node Labels and Node Selectors' e.g. kubectl label nodes node1 accelerator=example-gpu-x100, or requiredDuringSchedulingIgnoredDuringExecution node affinity on labels such as 'gpu.gpu-vendor.example/installed-memory' with operator Gt and NFD feature labels like 'feature.node.kubernetes.io/pci-10.present'. Node Feature Discovery 'detects the hardware features that are available on each node ... NFD can also add extended resources, annotations, and node taints. Administrators can leverage NFD to also taint nodes with specific features, so that only pods that request those features can be scheduled on those nodes.' Vendor implementations named: NVIDIA k8s-device-plugin and Intel device plugins. GPU example manifest uses resources.limits only, restartPolicy: OnFailure. Page notes items refer to third-party products and the Kubernetes project authors are not responsible for them.