All NVIDIA exams

NVIDIA NCP-AIN: NVIDIA-Certified Professional: AI Networking

What the NCP-AIN exam covers and what it costs, with a free 30-question practice test and an explanation for every answer.

At a glance
Exam fee
$400 USD
Duration
120 minutes
Passing score
Pass/fail
Level
Professional
Topic areas
5
Free practice questions
30
Full question bank
100 questions

What the NCP-AIN exam covers

The topic areas in our NCP-AIN question bank, with each one's share of the 100 practice questions.

NCP-AIN cost and format

Exam codeNCP-AIN
CertificationNVIDIA-Certified Professional: AI Networking
VendorNVIDIA
LevelProfessional
StatusActive
LanguagesEnglish

Last checked October 22, 2026 against the official NVIDIA exam page.

NCP-AIN sample questions

The first 5 questions from the free practice test. Open a question to see the answer and why it is right.

All 30 free questions
  1. Question 1Network telemetry and operations

    A training job runs slower than expected, and the team suspects a single congested uplink. Which metric should be checked first?

    1. AThe model's training accuracy and validation loss curve over the last 50 epochs
    2. BThe number of GPU CUDA cores and tensor cores installed on each compute node
    3. CThe switch's warranty expiry date and the vendor support contract end date
    4. DPer-port utilization and queue or buffer occupancy on the uplinks
    Show answer and explanation
    Correct answer: D

    Per-port utilization and queue occupancy show whether a specific uplink is congested. Training accuracy, GPU core counts and warranty dates do not identify a congested link.

  2. Question 2AI fabric fundamentals

    An AI training cluster runs large all-reduce operations across hundreds of GPUs. Which network characteristic matters most for the collective communication performance?

    1. ALow and predictable latency with high bandwidth and minimal packet loss
    2. BThe number of unrelated tenant services sharing the fabric
    3. CThe size and font of the switch's front-panel label printed on the chassis faceplate
    4. DMaximum routing table size for internet routes carried by the border routers
    Show answer and explanation
    Correct answer: A

    Collective operations are sensitive to latency variation and loss, because every GPU waits on the slowest transfer. Shared unrelated services, panel labels and internet route tables do not determine collective performance.

  3. Question 3InfiniBand and RoCE

    A cluster requires RDMA over Ethernet with lossless behaviour for AI traffic. Which technology provides RDMA over converged Ethernet?

    1. AICMP echo requests and replies used as a bulk data transfer path
    2. BStandard TCP over the kernel stack with no RDMA offload
    3. CRoCE, with congestion control and lossless configuration on the switches
    4. DARP address resolution broadcasts reused as a data transport
    Show answer and explanation
    Correct answer: C

    RoCE carries RDMA over Ethernet and requires lossless behaviour and congestion control, which are configured on the switches and NICs. Standard TCP, ICMP and ARP are not RDMA transports.

  4. Question 4Spectrum-X and Ethernet for AI

    A data centre designer wants Ethernet to carry AI traffic with performance close to a purpose-built fabric. Which approach is typically used with NVIDIA Spectrum-X?

    1. AStandard Ethernet switches with factory default settings, static ECMP hashing and no telemetry export
    2. BEthernet switches with adaptive routing and congestion control tuned for AI workloads
    3. CLegacy shared-medium Ethernet hubs connected in a daisy chain across the rack with half-duplex links
    4. DPoint-to-point serial connections between GPUs over RS-232 style links with no switching fabric
    Show answer and explanation
    Correct answer: B

    Spectrum-X combines Ethernet switches with adaptive routing and congestion control designed for AI traffic. Default switches lack that tuning, hubs cannot switch traffic at scale, and serial links do not provide cluster-scale fabrics.

  5. Question 5Security and multi-tenancy

    A security team needs to prevent a compromised GPU node from reaching other tenants' nodes on the fabric. Which control limits this lateral movement?

    1. AAllowing every node to reach each other freely across the fabric
    2. BAccess control lists or policies that restrict traffic between tenant segments
    3. CTurning off stateful firewalls on the fabric edge and tenant gateways during maintenance
    4. DGiving every node the shared administrator password and SSH key
    Show answer and explanation
    Correct answer: B

    ACLs and policies restrict traffic between segments and limit what a compromised node can reach. Open reachability, disabled firewalls and shared administrator passwords enable lateral movement.

How to prepare for NCP-AIN

  1. Read the official NVIDIA outline first (exam page), so you know exactly what is in scope.
  2. Give the most time to Network telemetry and operations and AI fabric fundamentals: together they make up 50% of the questions.
  3. Take the free 30-question practice test and note which topics you miss.
  4. Go back to the weak topics, then test again. A practice score shows where to study; it does not predict your exam result.

Full NCP-AIN question bank

All 100 questions with answers and explanations, as a PDF, an online practice exam, or both.

Full question bank · 100 questions
$89$124

Both formats together. The best value.

Instant access after payment, with 90 days of updates. Checkout is on our store, AIExamCerts.

NCP-AIN FAQ