NCA-AIIO Tests Matching NVIDIA Products to Purposes — Not Building Clusters by Hand
An entry-level, multiple-choice exam for people who plan, buy or support GPU infrastructure. No commands or output reading: spot the one keyword in each scenario and pick the product, interconnect, cooling or sharing model that fits.
Check Your Readiness →Most candidates understand NVIDIA-Certified Associate: AI Infrastructure and Operations (NCA-AIIO) concepts — and still fail. This exam tests how you apply knowledge under pressure.
The official 40% infrastructure domain is split in two here. Fifty short recognition items in 60 minutes, roughly a minute each. A scenario names a workload, constraint or lifecycle stage and you match it to a concept or NVIDIA product.
Add more GPUs inside each server
Idle time during cross-node communication points to the scale-out fabric. NVLink and NVSwitch are scale-up within a node or NVLink domain; InfiniBand or Spectrum-X Ethernet with GPUDirect RDMA connects servers.
Script nvidia-smi on each node and collect the output
nvidia-smi is a quick single-node view. DCGM provides telemetry, health checks and diagnostics at cluster scale and exports to Prometheus and Grafana. The BMC through IPMI or Redfish still works when the OS is down.
Replace the GPUs with a faster generation
The GPU is being starved by data loading, storage or the CPU. Fix the data path, for example with GPUDirect Storage or faster preprocessing such as RAPIDS; faster GPUs would simply wait longer.
Move it to on-demand cloud because cloud is cheaper
At sustained high utilization, owned on-premises infrastructure often costs less over time, and data egress adds cloud cost. Cloud suits fast starts and elastic bursts; hybrid balances control and flexibility.
TensorRT compiles a trained model for faster inference using layer fusion and reduced precision. Triton Inference Server hosts models from many frameworks with dynamic batching, and NIM packages an optimized model, runtime and standard API as a ready microservice.
CUDA is NVIDIA's parallel computing platform and programming model. PyTorch and TensorFlow are the frameworks built on it; cuDNN accelerates neural network layers and NCCL handles collective communication between GPUs across nodes.
RoCE delivers RDMA over Ethernet when the network is lossless or congestion-controlled, and Spectrum-X adds AI tuning such as adaptive routing. InfiniBand remains the established choice for dedicated training clusters, with native RDMA and in-switch SHARP reductions.
MIG partitions a GPU in hardware into up to seven instances with dedicated memory and compute. NVIDIA vGPU time-slices a GPU among VMs and needs licensing. Passthrough gives one VM the whole GPU at near bare-metal speed.
PUE is total facility power divided by IT equipment power, so values nearer 1.0 mean less overhead. AI racks now draw tens to over 100 kW, and the densest rack-scale systems need direct-to-chip liquid cooling.
Strengthen weak areas with exam-style practice questions and detailed explanations.
Simulate the real exam experience and assess your readiness under timed conditions.
Review key concepts, objectives, and exam topics in one place.
Get personalized explanations, learning recommendations, and instant answers.
Follow a structured learning path designed to help you prepare efficiently.
NCA-AIIO rewards picking the right NVIDIA product for each constraint. See how your matching holds up.