Fault-injection labs
22 guided labs walk through healthy baselines, injected failures, and recovery: ECC/XID triage, MIG partitioning, InfiniBand and RoCE fabrics, storage bottlenecks, Slurm and Kubernetes scheduling.
Aegis-GPU is a hands-on training simulator for NVIDIA's NCA-AIIO certification. Inject real fault signatures — XID errors, NVLink degradation, NCCL fallbacks, thermal throttling — into a virtual GPU fleet, then diagnose and recover them with the same commands you'd run in production.
22 guided labs walk through healthy baselines, injected failures, and recovery: ECC/XID triage, MIG partitioning, InfiniBand and RoCE fabrics, storage bottlenecks, Slurm and Kubernetes scheduling.
Type nvidia-smi, dcgmi, kubectl, and
scontrol against the simulated cluster. Correct diagnostics return
realistic output; weak ones get targeted coaching instead of a dead end.
113 practice questions across all three exam domains, weighted like the official blueprint. Every choice — right or wrong — gets specific feedback explaining the failure class behind it.
Provision a DGX-style fleet, watch live metrics respond to what you do, and work incidents across nodes, fabrics, and running jobs from a single operations view.
Scorecards track your reasoning per domain and tell you when your troubleshooting judgment is exam-ready — not just what percent you scored.
A built-in study guide maps every lab and quiz item to the official NCA-AIIO blueprint, so you can see exactly which objectives you still owe attention.