Skip to navigation

Kubernetes Operator Chart

View as MarkdownOpen in Claude

The nvidia-bnr-h4m-operator Helm chart installs the BNR Kubernetes operator and its NvidiaBnrMediaFunction custom resource definition (CRD). The operator reconciles a BNR NIM deployment and supporting resources from a custom resource.

The operator path is useful when you want the BNR media function to be managed declaratively through Kubernetes-native APIs.


Installation

Prerequisites

Complete all prerequisite steps (Rivermax license, image pull and model pull secrets, high-speed network attachment) before installing the operator. For details, refer to Getting Started.

Pull the Chart

helm pull nim-repo/nvidia-bnr-h4m-operator --version 2.0.0

For the full repository setup, refer to Pull Helm Charts.

Install the Operator

helm upgrade --install nvidia-bnr-h4m-operator \
nvidia-bnr-h4m-operator-2.0.0.tgz

With explicit image pull secret and image overrides:

helm upgrade --install nvidia-bnr-h4m-operator \
nvidia-bnr-h4m-operator-2.0.0.tgz \
--set imageCredentials.registry=nvcr.io \
--set imageCredentials.username='$oauthtoken' \
--set imageCredentials.password=$NGC_API_KEY

Check the operator:

kubectl get crd nvidiabnrmediafunctions.nvidia.com
kubectl get pods

On Red Hat OpenShift, replace kubectl with oc.


Create a BNR Media Function

Apply a NvidiaBnrMediaFunction custom resource after the operator is running.

NMOS Example

apiVersion: nvidia.com/v1alpha1
kind: NvidiaBnrMediaFunction
metadata:
name: nvidia-bnr-nim-media-function
spec:
schedulerName: topo-aware-scheduler
nodeSelector:
kubernetes.io/hostname: example-gpu-node
parameters:
highSpeedNetwork:
- name: media-a-tx-net
resources:
limits:
cpu: "4"
memory: "8Gi"
hugepages-2Mi: "2Gi"
nvidia.com/gpu: "1"
requests:
cpu: "4"
memory: "8Gi"
hugepages-2Mi: "2Gi"
nvidia.com/gpu: "1"
ngcSecret:
secretName: ngc-api-key
secretKey: NGC_API_KEY
nimModelCache:
enabled: true
size: "10Gi"
nimLogs:
enabled: true
size: "5Gi"
securityContext:
runAsUser: 1000
runAsGroup: 1000
runAsNonRoot: true
allowPrivilegeEscalation: true
capabilities:
add:
- IPC_LOCK
- NET_RAW
- SYS_NICE
- DAC_READ_SEARCH
drop:
- ALL
podSecurityContext:
fsGroup: 1000
seccompProfileType: RuntimeDefault
label: "BNR-NIM"
description: "BNR NIM - background noise removal, ST2110"
modelType: v2-48k
intensityRatio: "1.0"
nmosEnabled: true
nimEnabled: true
cpuAffinity: "0-3"
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
hostIP: "234.5.8.9"
hostPort: "5003"
hostNumSubnetBits: 24
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
hostIP: "234.5.8.9"
hostPort: "5004"
hostNumSubnetBits: 24

Create the custom resource:

kubectl apply -f nvidia-bnr-h4m.yaml

Check the reconciled workload:

kubectl get nvidiabnrmediafunction
kubectl get deploy,pods,pvc

Important Fields

FieldDescription
spec.schedulerNameScheduler used by the reconciled NIM pod. Defaults are typically set for topology-aware scheduling.
spec.nodeSelectorPins the NIM pod to the target GPU node.
spec.parameters.highSpeedNetwork[].nameMultus NetworkAttachmentDefinition for media traffic.
spec.parameters.ngcSecretGeneric secret containing the NGC API key used for model download.
spec.parameters.modelTypeBNR denoiser variant. Use v2-48k for 48 kHz input or v2-16k for 16 kHz input.
spec.parameters.intensityRatioDenoising strength from "0.0" to "1.0".
spec.parameters.labelNMOS label advertised by the BNR NIM when NMOS is enabled. The examples use BNR-NIM.
spec.inputs.audio_inputST 2110-30 audio input configuration.
spec.outputs.audio_outputST 2110-30 audio output configuration.

The sample_rate under inputs.audio_input.format.audio and outputs.audio_output.format.audio must match parameters.modelType: use 48000 with v2-48k, and 16000 with v2-16k.


Uninstall

Delete BNR media function resources before uninstalling the operator:

kubectl delete nvidiabnrmediafunction --all
helm uninstall nvidia-bnr-h4m-operator