Expose the configurable GPU power-limit range from GFD so NVIDIA Dynamo can consume it during hardware discovery.
Proposed labels:
- nvidia.com/gpu.power.limit.min
- nvidia.com/gpu.power.limit.max
GFD should query NVML GetPowerManagementLimitConstraints(). Values should be integers in milliwatts, matching NVML.
Current state: GPU Operator deploys GFD from k8s-device-plugin:v0.20.0. The vendored go-nvml binding already exposes the required API, but GFD’s device abstraction and labelers do not use it.
Acceptance criteria:
- Query minimum and maximum constraints for physical GPUs.
- Publish both labels through NodeFeature API and legacy feature-file output.
Use case: Dynamo needs the valid power-setting range to validate and optimize power configurations without querying NVML independently on every node.
Expose the configurable GPU power-limit range from GFD so NVIDIA Dynamo can consume it during hardware discovery.
Proposed labels:
GFD should query NVML GetPowerManagementLimitConstraints(). Values should be integers in milliwatts, matching NVML.
Current state: GPU Operator deploys GFD from k8s-device-plugin:v0.20.0. The vendored go-nvml binding already exposes the required API, but GFD’s device abstraction and labelers do not use it.
Acceptance criteria:
Use case: Dynamo needs the valid power-setting range to validate and optimize power configurations without querying NVML independently on every node.