Quick Start

This walkthrough takes you from a freshly prepared cluster to a running NVIDIA networking deployment using the SR-IOV Ethernet profile — the most common starting point. For other profiles, see the Deployment Profiles decision matrix.

Probe the cluster’s network hardware. Output is written to ./cluster-config.yaml by default.

            

            
l8k discover
        

Discovery deploys a minimal probe profile, examines NICs and OFED-dependent kernel modules on each node, and groups nodes by PCI topology.

Generate Kubernetes manifests for an SR-IOV Ethernet deployment. Launch Kit auto-reads ./cluster-config.yaml and writes manifests to ./deployment/.

            

            
l8k generate --fabric ethernet --deployment-type sriov --multirail
        

See Generate Workflow for the output file layout.

Apply the generated manifests in dependency order (NicClusterPolicy first, then per-group NicNodePolicy, then networks and workloads):

            

            
l8k deploy
        

Check that operator pods, OFED driver pods, and example workloads are running:

            

            
kubectl -n nvidia-network-operator get pods
kubectl get nicclusterpolicy
kubectl get nicnodepolicy
        

• Picking a different profile? Deployment Profiles

• Cluster has mixed GPU/NIC types? Heterogeneous Clusters

• Generating manifests without cluster access? Cluster Topology Presets

• Wiring l8k into CI/CD? Automation and CI/CD

• Something didn’t work? Troubleshooting

• Overview — the discover/generate/deploy mental model

• CLI Reference — every flag

• Configuration Reference — cluster-config.yaml schema