Skip to navigation

Operator Custom Resource

View as MarkdownOpen in Claude

The nvidia-bnr-h4m-operator chart installs the NvidiaBnrMediaFunction CRD in API group nvidia.com/v1alpha1.

Top-Level Spec

FieldDescription
spec.schedulerNameScheduler assigned to the reconciled BNR NIM pod.
spec.nodeSelectorKubernetes node selector map.
spec.parametersBNR runtime, image, secret, storage, resource, and security settings.
spec.inputs.audio_inputAudio input port configuration.
spec.outputs.audio_outputAudio output port configuration.

Parameters

FieldDescription
parameters.highSpeedNetwork[].nameMultus NetworkAttachmentDefinition used for media traffic.
parameters.highSpeedNetwork[].ipOptional static IP/CIDR requested from the network attachment.
parameters.resourcesResource requests and limits for the BNR NIM pod.
parameters.ngcSecret.secretNameGeneric secret containing the NGC API key.
parameters.ngcSecret.secretKeySecret key name. Default examples use NGC_API_KEY.
parameters.nimModelCachePVC configuration for downloaded model artifacts.
parameters.nimLogsPVC configuration for NIM logs.
parameters.securityContextContainer security context.
parameters.podSecurityContextPod security context.
parameters.labelNMOS label.
parameters.descriptionNMOS description.
parameters.modelTypeBNR model variant: v2-48k or v2-16k recommended.
parameters.intensityRatioDenoising strength as a string from "0.0" to "1.0".
parameters.nmosEnabledEnables NMOS runtime behavior.
parameters.nimEnabledEnables NIM runtime behavior.
parameters.cpuAffinityCPU affinity string passed to the runtime.
parameters.loggingLevelRuntime logging level from 0 to 5.
parameters.gstDebugOptional GST_DEBUG override.

Input and Output Ports

Each BNR media function has one audio input and one audio output:

inputs:
audio_input:
control: { nmos: {} }
transport:
rtp:
packet_time: 1
format:
audio:
sample_rate: 48000
channel_count: 1
sample_depth: 24
media_type: "audio/L24"
sessionName: audio_in
localInterfaceName: net1
outputs:
audio_output:
control: { nmos: {} }
transport:
rtp:
packet_time: 1
format:
audio:
sample_rate: 48000
channel_count: 1
sample_depth: 24
media_type: "audio/L24"
sessionName: audio_out
localInterfaceName: net1

Control Modes

Use exactly one control variant per input/output:

control: { nmos: {} }

Alternatively:

control:
sdp:
data: |
v=0
...

For NMOS deployments, use nmos. For fixed ST 2110 deployments, either use static host fields (hostIP, hostPort, hostNumSubnetBits) or inline SDP, depending on the operator workflow used by your cluster.

Model and Sample Rate

parameters.modelTypeformat.audio.sample_rate
v2-48k48000
v2-16k16000

The input and output sample rates must match the selected model variant.