Domain priorities and focus areas
Install and Deploy (32%) — highest weight:
- Know BCM installation steps
- Understand NGC container ecosystem (
nvcr.io,$oauthtoken) - Know DOCA architecture and BlueField deployment flow
- Understand storage tiers and GPUDirect Storage
Administration (28%) — second priority:
- MIG commands cold — especially
nvidia-smi mig -cgi,-cci,-lgi,-lci - Slurm job submission and node state management
- Fleet Command = edge, cloud service, no on-prem install
- Run:ai project/department/RBAC hierarchy
Workload Management (20%):
- kubectl commands for running and monitoring GPU workloads
- DCGM installation, modes, diagnostic levels
- BCM provisioning flow and cmsh commands
Troubleshooting (20%):
- Common Docker/NVML errors and their fixes
- Fabric Manager version must match driver version
- Fabric Manager = NVLink/NVSwitch; UFM = InfiniBand
