NCP-AIO AI Operations Professional Study Plan¶

8-Week Intensive Study Schedule¶

Phase 1: Foundation Building (Weeks 1-2)¶

Week 1: MLOps Fundamentals¶

Focus: ML lifecycle and pipeline automation

Day 1-2: ML Pipeline Architecture¶

• Study end-to-end ML pipeline stages
• Learn pipeline orchestration tools (Kubeflow, Airflow)
• Understand experiment tracking and reproducibility
• Reference: NVIDIA AI Enterprise

Day 3-4: Model Versioning and Registry¶

• Study model artifact management
• Learn model registry concepts and stage transitions
• Understand lineage tracking for reproducibility
• Practice with MLflow model registry

Day 5-7: CI/CD for ML¶

• Study ML-specific CI/CD practices
• Learn validation gates and promotion criteria
• Understand continuous training triggers
• Lab: Set up a basic ML pipeline with versioning

Week 2: Model Deployment¶

Focus: Triton and NIM deployment

Day 1-3: Triton Inference Server¶

• Study Triton architecture and model repository
• Learn config.pbtxt configuration
• Understand dynamic batching and model ensembles
• Practice deploying models on Triton
• Reference: Triton Documentation

Day 4-5: NVIDIA NIM¶

• Deploy NIM containers for LLM inference
• Compare NIM vs Triton capabilities
• Study auto-scaling configuration
• Reference: NIM Documentation

Day 6-7: Deployment Patterns¶

• Study blue-green, canary, and A/B testing patterns
• Learn rollback procedures
• Understand auto-scaling metrics and configuration
• Lab: Deploy a model with canary traffic splitting

Phase 2: Advanced Topics (Weeks 3-5)¶

Week 3: GPU Monitoring with DCGM¶

Day 1-2: DCGM Architecture¶

• Study DCGM components (engine, CLI, exporter)
• Learn health watch categories
• Understand diagnostic levels (1, 2, 3)
• Reference: DCGM Documentation

Day 3-4: Key Metrics¶

• Memorize important DCGM metrics and field IDs
• Learn ECC error types and their significance
• Study thermal and power monitoring
• Practice dcgmi commands

Day 5-7: Monitoring Stack¶

• Set up DCGM Exporter with Prometheus
• Create Grafana dashboards for GPU monitoring
• Configure alerting rules
• Lab: Build a complete GPU monitoring stack

Week 4: Fleet Management¶

Day 1-2: Multi-Cluster Operations¶

• Study fleet architecture patterns
• Learn centralized management approaches
• Understand inventory and capacity management

Day 3-4: Driver and Firmware Lifecycle¶

• Study driver update strategies
• Learn compatibility matrix management
• Understand firmware update procedures
• Reference: NVIDIA Driver Docs

Day 5-7: Configuration and Security¶

• Study Infrastructure as Code for GPU environments
• Learn security patching procedures
• Understand compliance requirements
• Lab: Automate GPU node configuration with Ansible

Week 5: Incident Response and Reliability¶

Day 1-2: GPU Failure Modes¶

• Study hardware failure types (ECC, NVLink, thermal)
• Memorize key Xid error codes
• Learn software failure patterns
• Practice diagnostic procedures

Day 3-4: Incident Response¶

• Study the 7-step incident response framework
• Practice troubleshooting scenarios
• Learn isolation and recovery procedures
• Lab: Simulate and respond to GPU incidents

Day 5-7: Disaster Recovery¶

• Study checkpointing strategies for training
• Learn HA patterns for inference services
• Understand RTO and RPO concepts
• Review SLA management practices

Phase 3: Review and Exam Prep (Weeks 6-8)¶

Week 6: Integration Practice¶

• End-to-end deployment scenario exercises
• Monitoring and incident response walkthroughs
• Fleet management simulations
• Practice multi-domain questions

Week 7: Review and Gap Analysis¶

• Review all five domains systematically
• Create flashcards for key metrics and commands
• Re-read documentation for weak areas
• Timed practice question sessions

Week 8: Final Preparation¶

• Review fact sheet and quick reference
• Full-length timed practice session
• Final review of weak areas
• Exam logistics preparation

Self-Assessment Questions¶

• Can I configure Triton model repository and dynamic batching?
• Do I know key DCGM metrics and diagnostic commands?
• Can I design a driver update strategy for a GPU fleet?
• Do I know Xid error codes and appropriate responses?
• Can I design disaster recovery for training and inference?