How does a pod request one GPU?
Why C is the answer
A resource limit of nvidia.com/gpu: 1
GPUs are extended resources, requested in limits.
In detail
- GPUs are extended resources, requested in limits.
- GPUs are extended resources.
Use case
An illustrative scenario, not a real incident.
A pod sets a limit of nvidia.com/gpu: 1, and it gets one GPU.
See it in different ways
flowchart TD
A["Pod Pending"] --> B["kubectl describe pod"]
B --> C{"Does the node report GPUs?"}
C -->|No| D["Check operator and device plugin"]
C -->|Yes| E["Check quotas, taints and selectors"]stateDiagram-v2 [*] --> Pending Pending --> NoGpuReported: node shows no GPUs Pending --> Taint: no matching toleration Pending --> WrongRequest: no nvidia.com/gpu limit NoGpuReported --> CheckOperator Taint --> AddToleration WrongRequest --> FixPodSpec
flowchart LR S0["Set the limit"] --> S1["Scheduler finds a node"] --> S2["Pod runs"]
Key point: Request GPUs in limits.
Why the other options are wrong
- A. An annotation naming the GPU UUID
- An annotation does not reserve a GPU.
- B. A ConfigMap with the driver version
- A ConfigMap does not reserve a GPU.
- D. A label reading gpu=yes on the pod
- A label does not reserve a GPU.