Reference on DRA Driver for NVIDIA GPUs

API reference

Mon, 01 Jan 0001 00:00:00 +0000

All types belong to API group resource.nvidia.com/v1beta1.

GPU opaque configuration types

These types are set as opaque configuration in ResourceClaim and ResourceClaimTemplate specs to configure GPU resources. They are optional, omit them to use driver defaults. Some fields required feature gates to be enabled before using.

GpuConfig

Configures a full GPU device. Target DeviceClass: gpu.nvidia.com.

With time-slicing:

apiVersion: resource.nvidia.com/v1beta1
kind: GpuConfig
sharing:
 strategy: TimeSlicing
 timeSlicingConfig:
 interval: Default  # Default | Short | Medium | Long

With MPS (Multi-Process Service):

Feature gates

Mon, 01 Jan 0001 00:00:00 +0000

Feature gates control experimental and beta functionality in the DRA Driver for NVIDIA GPUs. They follow Kubernetes feature gate conventions.

Set feature gates

Set feature gates in your Helm values file:

featureGates:
 TimeSlicingSettings: true
 MPSSupport: true

Or pass them at install time:

helm install nvidia-dra-driver-gpu nvidia/nvidia-dra-driver-gpu \
 --set "featureGates.TimeSlicingSettings=true"

Available feature gates

Feature gate	Stage	Default	Description
`TimeSlicingSettings`	Alpha	`false`	Enables customization of CUDA time-slicing settings in `GpuConfig`.
`MPSSupport`	Alpha	`false`	Enables Multi-Process Service (MPS) sharing strategy in `GpuConfig` and `MigDeviceConfig`.
`IMEXDaemonsWithDNSNames`	Beta	`true`	IMEX daemons use DNS names instead of raw IP addresses for peer communication. Required by `ComputeDomainCliques`.
`PassthroughSupport`	Alpha	`false`	Enables VFIO passthrough device allocation using `VfioDeviceConfig`.
`DynamicMIG`	Alpha	`false`	Enables dynamic MIG device allocation and reconfiguration. Kubernetes 1.33–1.35 requires `DRAPartitionableDevices` enabled on the kube-apiserver and kube-scheduler (see Kubernetes feature gates).
`NVMLDeviceHealthCheck`	Alpha	`false`	Enables GPU health checking using NVML.
`ComputeDomainCliques`	Beta	`true`	Uses `ComputeDomainClique` CRD objects to track IMEX daemon membership. Requires `IMEXDaemonsWithDNSNames`.
`CrashOnNVLinkFabricErrors`	Beta	`true`	Causes the kubelet plugin to crash rather than fall back to non-fabric mode when NVLink fabric errors are detected.
`DeviceMetadata`	Alpha	`false`	Enables IOMMU API device exposure (`/dev/iommu` or `/dev/vfio/vfio`) for VFIO workloads via `VfioDeviceConfig`. Requires `PassthroughSupport`.

Constraints

The following feature gate combinations are mutually exclusive and will cause a startup error:

Helm chart values

Mon, 01 Jan 0001 00:00:00 +0000

The dra-driver-nvidia-gpu Helm chart deploys the controller, kubelet plugin, optional admission webhook, and cluster-scoped DeviceClass resources. Set values in a custom values.yaml file or pass them with --set at install or upgrade time.

This page documents Helm values for the DRA Driver for NVIDIA GPUs. For install and upgrade steps, see Install and Upgrade. The authoritative source for defaults is values.yaml at that release.

To list the full set of values for the current release: