Flash sale 50% off
00d : 00h : 00m : 00s
aiaecel

NCP-AII Exam Guide: NVIDIA AI Infrastructure Professional

Everything you need to plan your NCP-AII preparation on one page: the exam format, the five domains and what each one tests, a 12-week study plan, and the lessons that cover every domain (the foundations are free).

Illustration of a GPU server

What is the NCP-AII certification?

NCP-AII (NVIDIA-Certified Professional: AI Infrastructure) is a professional-level exam for the people who build and run the infrastructure behind AI workloads. It checks that you can bring up GPU servers, validate a cluster from a single GPU to the whole fabric, install the control plane, and find and fix faults.

It is a scenario-heavy exam. Most questions describe a situation (a new node, a failing test, a slow job) and ask for the right next step, the right tool, or what a piece of output means. Memorizing definitions is not enough: you need to know the order of operations and what healthy output looks like.

Last reviewed October 2, 2026

The exam at a glance

Full name
NVIDIA-Certified Professional: AI Infrastructure
Exam code
NCP-AII
Level
Professional (intermediate)
Questions
70–75
Time
120 minutes (about 1.6 minutes a question)
Price
$400
Delivery
Online, remotely proctored

Exam details as published on the official NVIDIA certification page in September 2026. NVIDIA does not publish a passing score. Confirm the current details before you book: official NCP-AII page.

The five exam domains and their weights

The weights tell you where the points are. Test & Verification and System & Server Bring-up together are almost two thirds of the exam, so they deserve most of your study time.

Cluster Test & Verification

33%

Proving a cluster is ready: test each layer from one GPU up to the whole cluster and compare the results with a known-good baseline.

  • Single-node stress tests and HPL
  • Single-node NCCL tests, including NVLink Switch
  • Cable signal quality and correct cabling
  • Firmware and software on switches, BlueField-3 and transceivers
  • ClusterKit node assessment
  • NCCL east-west fabric bandwidth
  • NCCL, HPL and NeMo burn-in
  • Storage testing

System & Server Bring-up

31%

Taking servers from the loading dock to a validated, workload-ready state, in the right order.

  • The bring-up sequence and cluster topologies
  • BMC, out-of-band and TPM initial configuration
  • Firmware (including HGX) and fault detection
  • Power and cooling
  • Installing servers and GPUs
  • Cables and transceivers
  • Validating hardware for workloads
  • Initial parameters for third-party storage

Control Plane Installation & Configuration

19%

Installing and managing the software layer that turns servers into a cluster people can run jobs on.

  • Base Command Manager, including high availability
  • OS and cluster setup: categories, interfaces, Slurm, Enroot and Pyxis
  • Installing, updating and removing GPU and DOCA drivers
  • NVIDIA Container Toolkit and GPUs with Docker
  • NGC CLI

Troubleshoot & Optimize

12%

Finding the faulty part fast, replacing it safely, and getting full performance out of the hardware.

A 12-week NCP-AII study plan

Plan on 8–10 hours a week if you already work with Linux servers or networks. If you are new to data centers, add four to eight weeks at the start, or take the NCA-AIIO course first.

  1. 1

    Weeks 1–2: Foundations

    Linux for GPU servers, BMCs and out-of-band access, power and cabling, and AI networking. This is what every later domain assumes.

  2. 2

    Weeks 3–5: Bring-up (31%)

    Follow the bring-up sequence on paper until you can recite it: firmware, BMC, BIOS settings, GPU and topology checks. Keep a runbook of every command and its healthy output.

  3. 3

    Weeks 6–8: Test & Verification (33%)

    The biggest domain. Learn one tool per rung of the validation ladder (DCGM, NCCL tests, HPL, ClusterKit, storage tests, burn-in) and what a pass looks like for each.

  4. 4

    Weeks 9–10: Control plane (19%)

    Base Command Manager, software images and categories, the driver and CUDA version puzzle, Slurm, containers and Kubernetes with the GPU Operator.

  5. 5

    Week 11: Troubleshoot & physical layer (17%)

    A repeatable troubleshooting method, Xid errors, fabric and link faults, GPUDirect, MIG and BlueField.

  6. 6

    Week 12: Review and mock exam

    Learn the question patterns and the most-confused pairs, take a full timed mock exam, and revise your weakest domain from your runbook.

Do you need GPU hardware to prepare?

Not to start. The foundations (Linux, BMCs, networking) run on any Linux machine or VM, and control-plane tools such as Base Command Manager and Slurm can be practiced in local VMs.

For GPU-specific skills (nvidia-smi, DCGM diagnostics, NCCL tests, MIG), a few hours on NVIDIA LaunchPad or a rented multi-GPU cloud instance is enough. Our course also includes Cluster Lab, a browser simulator where you investigate realistic cluster faults without any hardware.

New to AI infrastructure? Start with NCA-AIIO

NCA-AIIO (AI Infrastructure and Operations) is NVIDIA's associate-level exam: 50 questions in one hour, aimed at beginners. Its material is the on-ramp to everything NCP-AII assumes. If terms like InfiniBand, BMC or MIG are new to you, start there.

NCP-AII exam FAQ

70 to 75 questions in 120 minutes, delivered online with a remote proctor. That is about 1.6 minutes per question, so pacing matters.

$400, as listed on the official NVIDIA certification page in September 2026. Check the official page for current pricing before you book.

NVIDIA does not publish a passing score. Prepare to be comfortable across all five domains rather than aiming for a number, and weight your study time by the domain percentages.

It is a professional-level, hands-on exam. Most questions are scenarios that ask for the next step, the right tool or the meaning of an output, so you need practical understanding of GPU servers, fabrics and cluster software, not just definitions. With some infrastructure background, 10–14 weeks of steady study is a realistic plan.

It is not required, but if you are new to AI infrastructure it is the easier on-ramp. NCA-AIIO covers the concepts (AI workloads, GPUs, networking, operations) that NCP-AII expects you to already know.

Cluster Test & Verification (33%) and System & Server Bring-up (31%). Together they are 64% of the exam.

No. Aiaecel is an independent learning platform. This guide and our courses are study material aligned with the public exam blueprint; they are not affiliated with, sponsored by or endorsed by NVIDIA. NVIDIA and NCP-AII are trademarks of NVIDIA Corporation.

Independent study material. Not affiliated with, sponsored by or endorsed by NVIDIA. NVIDIA, NCP-AII and NCA-AIIO are trademarks of NVIDIA Corporation, used here only to identify the exams.

Start your NCP‑AII prep today