Quick Start Guide for Kubernetes

Before You Begin

Before deploying the NVIDIA Network Operator, ensure you have the following:

Prerequisites

  1. Kubernetes Cluster: A running Kubernetes cluster (v1.19+) with nodes that have NVIDIA NICs.

  2. CLI Tools: Install kubectl and helm on your client machine:

    $ curl -fsSL -o get_helm.sh https://raw.githubusercontent.com/helm/helm/master/scripts/get-helm-3 \
        && chmod 700 get_helm.sh \
        && ./get_helm.sh
    
  3. Container Runtime: Nodes must be configured with a container engine such as CRI-O or containerd.

Install Network Operator Helm Chart

Add the NVIDIA NGC Helm repository:

helm repo add nvidia https://helm.ngc.nvidia.com/nvidia
helm repo update

Install the Network Operator:

helm install network-operator nvidia/network-operator \
  -n nvidia-network-operator \
  --create-namespace \
  --version v25.7.0-rc.1 \
  --set sriovNetworkOperator.enabled=true \
  --wait

Verify the installation:

kubectl -n nvidia-network-operator get pods

Overview of Quickstart Use Cases

This quick start guide covers five essential networking configurations for different computational requirements:

Use Case

Purpose

Performance Requirements

Applications

SR-IOV Network with RDMA

High-performance networking with hardware acceleration

  • >10 Gbps throughput

  • <1μs latency

  • Dedicated VF resources

HPC simulations, distributed ML training, financial trading

Keywords: SR-IOV, RDMA, HPC, low-latency, VF isolation

Host Device Network with RDMA

Direct hardware access for legacy applications

  • Raw device control

  • Exclusive hardware access

  • Minimal CPU overhead

Legacy HPC codes, specialized protocols, DPDK applications

Keywords: host-device, PCI-passthrough, direct-access, exclusive-access

IP over InfiniBand with RDMA Shared Device

InfiniBand networking with shared RDMA resources

  • >50 Gbps bandwidth

  • Parallel I/O workloads

  • Shared device efficiency

Distributed storage, data analytics, scientific computing

Keywords: InfiniBand, IPoIB, shared-device, high-bandwidth

MacVLAN Network with RDMA Shared Device

Network isolation with shared RDMA capabilities

  • Multi-tenant segmentation

  • 10+ pods per node

  • Moderate throughput

Cloud-native HPC, microservices, multi-tenant ML

Keywords: MacVLAN, multi-tenant, network-segmentation, resource-sharing

SR-IOV InfiniBand Network with RDMA

Virtualized InfiniBand with hardware acceleration

  • >100 Gbps bandwidth

  • Hardware acceleration

  • Isolated IB partitions

Large-scale HPC clusters, AI/ML training, research computing

Keywords: SR-IOV, InfiniBand, hardware-acceleration, ultra-high-bandwidth