Generative AI
Building with large language models, multimodal models and agents — prompting, retrieval, fine-tuning, evaluation and serving. This is the track for application and ML engineers.
All 12 NVIDIA certifications, and which one to sit
NVIDIA offers 12 certifications across four tracks — Generative AI, AI Infrastructure, Data Science, and Simulation and Physical AI. 4 are associate-level (one hour, $125) and 8 are professional-level (two hours, $200 to $500). Every exam is delivered online under remote proctoring, is valid for two years, and has no prerequisite certification — you may sit any of them in any order.
Every figure on this page is transcribed from NVIDIA’s own certification pages and was last checked on . Prices and policies are set by NVIDIA and can change — verify on nvidia.com before booking. passgenai is an independent study resource and is not affiliated with, endorsed by, or sponsored by NVIDIA Corporation.
Every exam at a glance
The whole programme on one screen, cheapest first within each track. Sitting all 12 once would cost $3,000, which is a reason to choose deliberately rather than collect.
| Exam | Level | Fee | Length | Questions | Pass mark | Prep |
|---|---|---|---|---|---|---|
| NCA-GENLNVIDIA Certified Associate: Generative AI LLMs | Associate | $125 | 1 hour | 50–60 | Not published | Free exam guide → |
| NCA-GENMNVIDIA Certified Associate: Generative AI Multimodal | Associate | $125 | 1 hour | 50–60 | Not published | Free exam guide → |
| NCP-GENLNVIDIA Certified Professional: Generative AI LLMs | Professional | $200 | 2 hours | 60–70 | 70% | Free exam guide → |
| NCP-AAINVIDIA Certified Professional: Agentic AI | Professional | $200 | 2 hours | 60–70 | 70% | Free exam guide → |
| NCA-AIIONVIDIA-Certified Associate: AI Infrastructure and Operations | Associate | $125 | 1 hour | 50 | Not published | NVIDIA page ↗ |
| NCP-AIINVIDIA-Certified Professional: AI Infrastructure | Professional | $400 | 2 hours | 40–60 (not published per exam) | Not published | NVIDIA page ↗ |
| NCP-AIONVIDIA-Certified Professional: AI Operations | Professional | $500 | 2 hours | 40–60 (not published per exam) | Not published | NVIDIA page ↗ |
| NCP-AINNVIDIA-Certified Professional: AI Networking | Professional | $400 | 2 hours | 40–60 (not published per exam) | Not published | NVIDIA page ↗ |
| NCP-ARINVIDIA-Certified Professional: AI Rack and Interconnect | Professional | $400 | 2 hours | 40–60 (not published per exam) | Not published | NVIDIA page ↗ |
| NCA-ADSNVIDIA-Certified Associate: Accelerated Data Science | Associate | $125 | 1 hour | 40–60 (not published per exam) | Not published | NVIDIA page ↗ |
| NCP-ADSNVIDIA-Certified Professional: Accelerated Data Science | Professional | $200 | 2 hours | 40–60 (not published per exam) | Not published | NVIDIA page ↗ |
| NCP-OUSDNVIDIA-Certified Professional: OpenUSD Development | Professional | $200 | 2 hours | 40–60 (not published per exam) | Not published | NVIDIA page ↗ |
This site publishes free exam guides, study plans, prep courses, cheatsheets and glossaries for 4 of these 12 exams. For the rest, the row links to NVIDIA’s own page — we would rather send you to the vendor than pretend to cover an exam we do not.
The four tracks, and who each is for
Your track is decided by what you do, not by which exam looks most impressive. Pick the track first and the exam within it second.
AI Infrastructure
Specifying, building and running the GPU estate underneath the models — servers, networking, racks, power, cooling, orchestration and monitoring. This is the track for infrastructure, platform and data-centre engineers.
- NCA-AIIOAssociate$125
- NCP-AIIProfessional$400
- NCP-AIOProfessional$500
- NCP-AINProfessional$400
- NCP-ARIProfessional$400
Data Science
GPU-accelerated data science: moving dataframe, machine-learning and graph workloads onto the GPU with the RAPIDS stack instead of scaling out on CPUs.
- NCA-ADSAssociate$125
- NCP-ADSProfessional$200
Simulation and Physical AI
Describing and composing 3D worlds with OpenUSD — the format underneath digital twins, simulation and robotics pipelines.
- NCP-OUSDProfessional$200
What each exam actually covers
One paragraph per exam, written from the published domain weights — so you can tell NCA-GENL from NCA-GENM without opening four vendor tabs.
NCA-GENL — NVIDIA Certified Associate: Generative AI LLMs
Machine-learning and transformer fundamentals, prompt engineering, retrieval-augmented generation, Python NLP tooling, experimentation and evaluation, and trustworthy AI — five domains, weighted toward core ML at 30%.
Who sits it: Developers who build with LLMs and want a first credential. The usual entry point to the whole programme.
Assumed background: A basic understanding of generative AI and large language models
NCA-GENM — NVIDIA Certified Associate: Generative AI Multimodal
The same associate level as NCA-GENL but spread across seven domains covering image, audio and video alongside text, with experimentation the heaviest at 25%.
Who sits it: Developers whose work involves vision or speech models as much as language ones.
Assumed background: A basic understanding of generative AI
NCP-GENL — NVIDIA Certified Professional: Generative AI LLMs
Ten domains weighted toward model optimization (17%), GPU acceleration (14%), prompt engineering and fine-tuning (13% each) — the engineering of making an LLM fast, adapted and reliable in production.
Who sits it: Engineers who already architect and operate LLM systems and want the credential that reflects it.
Assumed background: 2–3 years of practical AI/ML experience with large language models, including transformer architectures, prompt engineering, distributed parallelism and parameter-efficient fine-tuning
NCP-AAI — NVIDIA Certified Professional: Agentic AI
Ten domains on agents specifically: architecture and development (15% each), evaluation and tuning, deployment and scaling (13% each), then cognition and memory, knowledge integration, platform tooling, operations, safety and human oversight.
Who sits it: Engineers building multi-agent and tool-using systems who already have the LLM fundamentals.
Assumed background: 1–2 years in AI/ML roles with hands-on work on production agentic AI projects — agent architecture, orchestration, multi-agent frameworks, evaluation, observability and guardrails
NCA-AIIO — NVIDIA-Certified Associate: AI Infrastructure and Operations
Three domains: AI infrastructure at 40% (GPU sizing and scaling, power, cooling, networking, on-prem versus cloud), essential AI knowledge at 38%, and AI operations at 22% (cluster orchestration, GPU monitoring, virtualization).
Who sits it: Infrastructure and platform engineers, and anyone specifying or buying GPU capacity. The cheapest way into the infrastructure track.
Assumed background: A basic understanding of data center infrastructure
NCP-AII — NVIDIA-Certified Professional: AI Infrastructure
Designing and deploying GPU infrastructure for AI workloads at data-centre scale.
Who sits it: Infrastructure engineers who build AI clusters rather than consume them.
NCP-AIO — NVIDIA-Certified Professional: AI Operations
Running AI infrastructure in production — orchestration, monitoring, capacity and fault management across a GPU estate.
Who sits it: Platform and SRE teams operating GPU clusters. The most expensive exam in the programme at $500.
NCP-AIN — NVIDIA-Certified Professional: AI Networking
The network fabric AI training depends on — InfiniBand and Ethernet for collective communication, topology, congestion and tuning.
Who sits it: Network engineers responsible for the interconnect in a training cluster.
NCP-ARI — NVIDIA-Certified Professional: AI Rack and Interconnect
Rack-scale integration — physical build, interconnect, power and cooling for dense GPU systems.
Who sits it: Engineers who deploy rack-scale GPU systems on the data-centre floor. The newest corner of the programme.
NCA-ADS — NVIDIA-Certified Associate: Accelerated Data Science
GPU-accelerated data science fundamentals — the RAPIDS stack (cuDF, cuML, cuGraph) and when moving a workload to the GPU pays for itself.
Who sits it: Data scientists and analysts whose datasets have outgrown pandas on a CPU.
NCP-ADS — NVIDIA-Certified Professional: Accelerated Data Science
Building and tuning accelerated data-science pipelines end to end, including multi-GPU and out-of-core work.
Who sits it: Data scientists and ML engineers already working on GPUs day to day.
NCP-OUSD — NVIDIA-Certified Professional: OpenUSD Development
Developing with OpenUSD — composition, layering, schemas and the pipelines behind digital twins and simulation.
Who sits it: 3D, simulation and robotics developers working on digital twins or Omniverse pipelines.
Is there a required order?
No — but there is a sensible one.
There is no mandatory sequence — every NVIDIA exam is open to anyone, and no certification is a prerequisite for another. What exists is a sensible order, and it is decided by your track rather than by the programme. Pick the track that matches your work: Generative AI if you build with models, AI Infrastructure if you run the GPUs underneath them, Data Science if your problem is dataframe and machine-learning workloads at scale, Simulation if you work in OpenUSD and digital twins. Then sit that track's associate exam first ($125, one hour) and its professional exam once you have real delivery experience — the professional papers assume one to three years of hands-on work and are written for someone who has operated the systems they describe, so taking one early is an expensive way to discover that. The one genuine dependency is conceptual: NCP-AAI builds on LLM fundamentals, so NCA-GENL or equivalent experience should come first.
The rules that apply to every exam
Programme-wide policy, stated once. These are NVIDIA's terms, not ours.
- Validity
- Two years
- NVIDIA certifications are valid for two years, after which you retake the exam to recertify. There is no continuing-education path, so budget the fee again in two years or treat the credential as a snapshot.
- Retake waiting period
- 14 days
- A failed attempt cannot be retaken for 14 days. That is long enough to rebuild a weak domain properly, and the score report tells you which one — so the wait is a feature rather than a delay.
- Attempt limit
- Five per exam per year
- You may not take any given exam more than five times in a year. In practice nobody reaches this, but it does rule out brute-forcing the question pool.
- Question count
- 40–60 for most exams
- NVIDIA states most of its exams contain between 40 and 60 questions. Where an exam page publishes its own figure — 50 for NCA-AIIO, 60–70 for the professional generative-AI exams — that number is the one to pace against.
- Cancellation
- Non-refundable within 24 hours
- Exams are non-cancellable and non-refundable inside 24 hours of the scheduled session. Reschedule the day before if you are not ready; on the day, the fee is gone whether you sit it or not.
- Delivery
- Online, remotely proctored
- No test centre. You need a webcam, a stable connection, government photo ID and a room you can clear of notes and second screens — the proctor will ask you to show it.
Common questions
What is the NVIDIA certification path?
There is no mandatory sequence — every NVIDIA exam is open to anyone, and no certification is a prerequisite for another. What exists is a sensible order, and it is decided by your track rather than by the programme. Pick the track that matches your work: Generative AI if you build with models, AI Infrastructure if you run the GPUs underneath them, Data Science if your problem is dataframe and machine-learning workloads at scale, Simulation if you work in OpenUSD and digital twins. Then sit that track's associate exam first ($125, one hour) and its professional exam once you have real delivery experience — the professional papers assume one to three years of hands-on work and are written for someone who has operated the systems they describe, so taking one early is an expensive way to discover that. The one genuine dependency is conceptual: NCP-AAI builds on LLM fundamentals, so NCA-GENL or equivalent experience should come first.
Which NVIDIA certification should you take first?
For most people, NCA-GENL — the associate generative-AI exam. It is the cheapest tier at $125, assumes only a basic understanding of LLMs, and its syllabus is the vocabulary the rest of the programme takes for granted. Two exceptions worth naming: if your job is infrastructure rather than application work, start with NCA-AIIO instead, and if you work mainly with images, audio or video, NCA-GENM is the closer match. Do not start with a professional exam.
How many NVIDIA certifications are there?
There are 12 NVIDIA certifications, across four tracks: four in Generative AI (NCA-GENL, NCA-GENM, NCP-GENL, NCP-AAI), five in AI Infrastructure (NCA-AIIO, NCP-AII, NCP-AIO, NCP-AIN, NCP-ARI), two in Data Science (NCA-ADS, NCP-ADS) and one in Simulation and Physical AI (NCP-OUSD). Four are associate-level and eight professional.
Is there an NVIDIA certification for beginners?
The associate tier is the beginner tier, and its prerequisites are deliberately light — NCA-GENL asks for a basic understanding of generative AI and LLMs, NCA-AIIO for a basic understanding of data-centre infrastructure. Neither expects professional experience. They are not trivial exams: they cover a broad syllabus in an hour and reward judgement over recall. But a motivated beginner with a few weeks of structured study can pass one, which is not true of any professional exam in the programme.