You want Kubernetes nodes to get GPU drivers, the device plugin and monitoring automatically. What do you install?
Why B is the answer
The NVIDIA GPU Operator
The GPU Operator automates the software needed for GPUs on Kubernetes nodes.
In detail
- The GPU Operator automates drivers, the device plugin and monitoring.
- It covers the software GPUs need on nodes.
Use case
An illustrative scenario, not a real incident.
A team adds GPU nodes to Kubernetes every month. The GPU Operator sets each one up, so nobody installs parts by hand.
See it in different ways
flowchart TD A["GPU Operator"] --> B["Driver"] A --> C["Container toolkit"] A --> D["Device plugin"] A --> E["DCGM exporter"] A --> F["MIG manager"]
mindmap
root((GPU Operator))
Driver
Container toolkit
Device plugin
Reports GPUs
DCGM exporter
Publishes metrics
MIG managerflowchart LR S0["Node added"] --> S1["Operator installs driver"] --> S2["Device plugin starts"] --> S3["Monitoring runs"]
Key point: GPU Operator automates node software.
Why the other options are wrong
- A. A static pod per node by hand
- Hand-built pods are hard to maintain.
- C. The NVIDIA Container Toolkit alone
- The toolkit alone covers only container access.
- D. The Slurm controller
- Slurm is a separate scheduler.