NCA-AIIO AI Infrastructure Operations Associate Study Strategy¶

Study Approach¶

Phase 1: Foundation (1-2 weeks)¶

1. GPU Monitoring - nvidia-smi, DCGM, health indicators
2. Containers - Container Toolkit, Docker GPU flags, NGC
3. 📖 DCGM Docs
4. 📖 Container Toolkit Docs

Phase 2: Operations (2-3 weeks)¶

1. Kubernetes - GPU Operator, scheduling, MIG
2. Infrastructure - Drivers, CUDA compatibility, maintenance
3. Troubleshooting - Common issues, Xid errors, log analysis
4. 📖 GPU Operator Docs

Phase 3: Exam Prep (1 week)¶

1. Practice scenario-based questions
2. Review commands and error codes
3. Focus on hands-on procedures

Recommended Resources¶

• DCGM Documentation - GPU monitoring
• Container Toolkit - GPU containers
• GPU Operator - Kubernetes GPU
• NGC Catalog - Container registry
• NVIDIA DLI Courses - Official training

Exam Tactics¶

Keywords¶

• "Monitor" or "health" - nvidia-smi or DCGM
• "Container" or "Docker" - Container Toolkit, --gpus flag
• "Kubernetes" or "pod" - GPU Operator, scheduling
• "Error" or "failure" - Xid codes, troubleshooting steps
• "Isolation" - MIG
• "Share GPU" - MIG or time-slicing
• "Driver" - Version compatibility, update procedure

Common Pitfalls¶

• nvidia-smi shows driver info; nvcc shows CUDA toolkit version
• Container Toolkit enables GPU containers; GPU Operator manages Kubernetes GPU stack
• SBE is correctable and normal at low rates; DBE is always critical
• MIG provides hardware isolation; time-slicing provides no memory isolation
• GPU memory is separate from system RAM
• CUDA runtime version in containers must be compatible with host driver

Time Management¶

• 60 minutes for 50-60 questions
• ~1 minute per question
• Flag uncertain questions and return at end
• Do not overthink straightforward command questions