GPU workloads
Task-oriented walkthroughs for running GPU workloads.
Guides
Task-oriented walkthroughs for common workflows.
These guides walk through how to use specific features of the driver. For
example manifests, see
demo/
in the repository.
ComputeDomain workloads
Create a ComputeDomain, claim a channel, and run a Multi-Node NVLink workload.
GPU sharing with Multi-Process Service
Share a single GPU between multiple containers using NVIDIA Multi-Process Service (MPS).
GPU health checking
Monitor GPU health using NVML and apply device taints to prevent new workloads from scheduling on unhealthy GPUs.
Troubleshooting
Collect diagnostic data and tune log verbosity when something goes wrong.