NVIDIA Certified Associate – AI Infrastructure & Operations (NCA-AIIO)

Learning plan

About 32 hours of focused study, split into 4 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 4 weeks

What the exam covers

Source: official exam guide. Weights can change, so check before you start.

Before you start

• NVIDIA's stated prerequisite: a basic understanding of data centre infrastructure
• Linux command-line basics and some exposure to containers
• A Certiverse account to book the exam

1. Week 1 · 9 h

   Essential AI knowledge

   • Tell AI, ML, deep learning and generative AI apart, and contrast training with inference workloads
   • Explain why GPUs beat CPUs for parallel matrix maths, and what Tensor Cores and reduced precision (FP16, BF16, FP8) add
   • Map the NVIDIA software stack (CUDA, cuDNN, NGC containers, NVIDIA AI Enterprise, Triton, NIM) to what each does

   Use: NVIDIA AI Infrastructure and Operations Fundamentals (self-paced) · NCA-AIIO exam study guide (PDF) · NVIDIA NGC catalog

2. Weeks 1–2 · 11 h

   AI infrastructure (largest domain)

   • Describe DGX/HGX systems, NVLink and NVSwitch for GPU-to-GPU traffic, and scale-up vs scale-out
   • Compare InfiniBand with Ethernet/RoCE for east-west AI traffic, and explain what BlueField DPUs offload
   • Size storage for training (parallel file systems, GPUDirect Storage) and plan power, cooling and rack density for GPU clusters
   • Choose between on-premises, cloud and hybrid AI infrastructure by cost and utilisation

   Hands-on: Rent a cloud GPU VM (or use a free Colab/Kaggle GPU) and run nvidia-smi and nvidia-smi topo -m. Write down GPU model, memory, driver/CUDA versions and interconnect, then run an NGC PyTorch container with the NVIDIA Container Toolkit.

   Use: NVIDIA networking documentation · NVIDIA Multi-Instance GPU (MIG) user guide

3. Week 3 · 7 h

   AI operations

   • Monitor GPU health and utilisation with DCGM and export metrics to Prometheus/Grafana
   • Run GPU workloads on Kubernetes with the GPU Operator and on HPC clusters with Slurm, and know when each scheduler fits
   • Partition GPUs with MIG and know when time-slicing or vGPU fits better

   Hands-on: On a single-GPU cloud VM, install k3s and the NVIDIA GPU Operator, schedule a pod that requests nvidia.com/gpu, and scrape its utilisation with the DCGM exporter while the job runs.

   Use: NVIDIA DCGM documentation · NVIDIA GPU Operator documentation · NVIDIA Container Toolkit documentation

4. Weeks 3–4 · 5 h

   Practice and review

   • Work through NVIDIA's practice questions in learning mode and read every explanation, including for correct answers
   • Re-study AI Infrastructure (40%) and Essential AI Knowledge (38%), which make up 78% of the exam

   Use: NCA-AIIO exam page

You're ready to book when…

• You can explain NVLink vs InfiniBand vs Ethernet in one sentence each
• You have run a GPU container and a Kubernetes GPU pod yourself and read DCGM metrics
• You can map each NVIDIA software component to its job without notes
• You score 80% or higher on the official practice questions

Exam tips

• The exam is 50 questions in 60 minutes, online and remotely proctored through Certiverse. It is valid for two years and you renew by retaking it.
• NVIDIA's exam page offers practice questions in a learning mode with immediate feedback and explanations. Use them before booking.