Durga Analytics training · instructor-led

Certified Professional AI Networking Training

Sixty percent of the blueprint splits evenly between Spectrum Ethernet and InfiniBand, so you cannot pass this knowing only the fabric you happen to run.

Six modules mapped to the published blueprint, taught live by practitioners, and built so the capability outlasts the exam. Take it as a scheduled cohort or have it delivered privately to your team.

DA-400 Specialist · Network Engineer · 6 modules · cohort or private team delivery

Or take the wider track

This exam is covered by NVIDIA AI Infrastructure Mastery

If you want the platform rather than the certificate, the mastery track covers everything this exam tests and a good deal it does not. Our programme leads to the certification, not the other way round. Take this page's programme when you need the credential on a date; take the track when you need the capability.

Start here · no exam required · DA-200 to DA-400

NVIDIA AI Infrastructure Mastery

GPU estates fail on networking, scheduling and thermals long before they fail on model code. A working command of GPU infrastructure and the NVIDIA software stack, covering the ground the NCA and NCP credentials examine.

The programme

What it covers

Six modules mapped to the published exam blueprint, each ending in applied work rather than a quiz. The schedule is agreed before the programme starts.

  1. AI data centre design

    Rail-optimised and fat-tree topologies, oversubscription choices, and sizing a fabric against the collective pattern the training job will run.

  2. Spectrum Ethernet

    Cumulus Linux and NVUE, RoCE with PFC and ECN, adaptive routing, and the tuning that stops a lossless fabric from stalling.

  3. InfiniBand

    Subnet manager and OpenSM, partitioning with pkeys, SHARP in-network aggregation, adaptive routing, and congestion control on the IB fabric.

  4. Troubleshooting tools

    UFM, ibdiagnet, link and symbol error counters, NCCL tests used as a fabric probe, and finding the one bad transceiver.

  5. Kubernetes and automation

    Network Operator, SR-IOV and multus for GPU pods, plus NVUE automation and Ansible-driven switch configuration at fleet scale.

  6. Exam drilling

    Two-hour format with seventy to seventy-five questions, pass or fail with no score reported, and timed full-length practice.

How it is delivered

Two ways to take it

Both tiers are live, instructor-led delivery. Schedule and depth are agreed before the programme starts rather than fixed in advance.
FeatureCohortEnterprise
FormatScheduled live sessions with a mixed peer groupThe same programme delivered privately to your team
Live sessionsInstructor-led, scheduledInstructor-led, scheduled around your delivery calendar
LabsShared lab scenariosRun against your own estate where access allows
TailoringFixed syllabusSequenced to your platforms, workloads and governance
Exam focusBlueprint coverage plus question-format drillingBlueprint coverage, plus the gaps found in your team's baseline
Group sizeIndividual seatsWhole team, one cohort
Best forOne person preparing around a jobA team standardising on the same platform
Reference

The certification itself

The credential this programme prepares you for, as NVIDIA publishes it. Open the official page from the exam code before you book or budget.

Level
Exam fee
400 USD
Validity
2 years

The exam fee is paid to NVIDIA, not to us, and is separate from the programme. See the NVIDIA credential reference for the rest of the ladder and the renewal rules.

Take this programme

Register and we will come back with cohort dates, scope and pricing. For a team, tell us how many people and which platforms you run and we will scope a private delivery.

Independent training. Durga Analytics is not a vendor partner and does not award credentials. Verify exam details officially. We prepare people for the exam; no training provider can guarantee a pass. Credential detail reviewed 15 October 2026.