NCP-AII practice questions
NVIDIA · NCP-AII · 200 questions
Original practice questions for the NVIDIA-Certified Professional: AI Infrastructure (NCP-AII) exam, covering system and server bring-up, physical layer management (BlueField and MIG), control plane installation and configuration (BCM, Slurm/Enroot/Pyxis, drivers, NVIDIA Container Toolkit, NGC CLI), cluster test and verification (HPL, NCCL, ClusterKit, burn-in, storage), and troubleshoot and optimize — framed in civic and municipal scenarios such as city AI data centers, transit authority GPU clusters, public-health inference racks, and school-district AI labs.
This course contains the use of artificial intelligence.
About the NCP-AII exam
- Time allowed
- 2 hours
- Questions
- 70-75
- Languages
- English
Exam details published by the vendor, checked 6 November 2026. Vendors change fees and formats without notice — confirm on the vendor's own page before you book.
Practice Quizzes
Test your knowledge with standard 20-question practice sets.
Quiz 1
Quiz 2
Quiz 3
Quiz 4
Quiz 5
Quiz 6
Quiz 7
Quiz 8
Quiz 9
Quiz 10
Browse by Domain
Study specific topics at your own pace.
System and Server Bring-up · 62 questions
- A city AI factory racks new HGX nodes for a permitting vision model. Which day-0 order should the rack team follow before any workload starts?
- County public-health IT is writing a checklist for new inference nodes. Which validation gates belong before control-plane install and Slurm jobs?
- A transit authority signs off a SuperPOD-style row and asks where cable and transceiver verification sits relative to BMC setup and GPU install. What is the correct placement?
- An emergency-management compute lead debates when third-party storage initial parameters should be set during AI node bring-up. What ordering is appropriate?
- A school-district AI lab's first rack failed because teams installed the OS before power and cooling validation. What risk does that out-of-order sequence create?
- A housing-authority analytics cluster runbook validates nvidia-smi health only after firmware upgrades finish. Why is that sequencing correct?
- City data-center architects compare fabrics for multi-node training of 311 NLP models with heavy east-west GPU traffic. Which topology choice best fits an AI factory pattern?
- A county assessor training cluster needs non-blocking paths between HGX nodes for all-reduce heavy jobs. Which topology property should planners prioritize?
- Transit operations separates management out-of-band networking from the GPU compute fabric in a municipal AI row. What distinction should the design preserve?
- A public library consortium asks why AI factories often use rail or fat-tree patterns instead of only a classic enterprise three-tier LAN. What is the best explanation?
- A water-utility digital-twin rack plan must relate storage and GPU networks. Which description fits AI factory bring-up design?
- Courts e-filing OCR GPUs scale from one rack to multiple. When should planners expect an AI-factory topology redesign?
- A municipal rack team is bringing up new DGX-class nodes for city services. What should they do with BMC before any in-band OS install?
- County IT insists TPM is initialized and policy-aligned before public-health GPUs enter production. What bring-up outcome does that support?
- The city NOC cannot reach a newly racked GPU node in-band and needs first console access. Which approach is correct?
- A school-district audit finds AI lab BMCs still on default passwords. What initial BMC hardening step should operators take?
- A transit authority maps dedicated OOB switches so GPU nodes stay manageable during fabric outages. What configuration goal does that serve?
- A housing authority verifies TPM enrollment status through BMC tools before accepting a node into the municipal cluster. What gate is being enforced?
- A city upgrades HGX firmware on a permitting-model rack. What should operators do regarding faults before returning nodes to Slurm?
- County public-health operators stage firmware bundles for GPU servers. What coordination practice is most appropriate?
- Transit ops detects a failed firmware step on one HGX tray during a row upgrade. What is the safest immediate response?
- Emergency management needs 911 NLP inference capacity to stay N+1 during firmware windows. How should upgrades be planned?
- A school-district AI lab follows vendor guidance on HGX firmware versus driver updates. Which ordering principle is appropriate?
- A courts OCR cluster re-runs health checks after firmware upgrades. What problem does that practice catch?
- City facilities must clear a dense municipal GPU row for power-on. Which validation should happen first against the AI server requirements?
- A county AI room alarms on inlet temperature during a burn-in rehearsal for health-inference GPUs. What should operators do first for power and cooling validation?
- A transit authority must power multi-PSU HGX nodes for a municipal training row. Which power validation best supports redundancy policy?
- A water-utility digital-twin rack is about to accept new AI nodes. What cooling check belongs in bring-up before sustained load?
- A housing-authority analytics cluster measures rack draw above designed PDU headroom during AI bring-up. What is the correct response?
- A library consortium installs grant-funded GPU servers and must document power for facilities. Which practice validates power parameters at install?
- City administrators finish OS and driver install on new permitting GPU nodes. What SMI-centered check confirms the servers are ready?
- County public-health operators must accept new inference hosts into the queue. Which SMI inventory check is an appropriate install gate?
- Transit operations wants every new HGX host checked the same way during municipal install. How should nvidia-smi fit the workflow?
- A school-district AI lab racks a new GPU server but nvidia-smi reports zero GPUs. What is the best next diagnostic move at install?
- Emergency-management standby inference hosts need known GPU operational settings after install. What should operators validate with nvidia-smi?
- A courts OCR farm accepts new GPU servers into municipal inventory. How should nvidia-smi support asset tracking at install?
- City AI-factory operators finish racking a municipal row and must prove hardware completeness. Which installed-hardware validation fits day-0?
- County public-health IT compares a GPU server’s bill of materials to live discovery before sign-off. What outcome should block acceptance?
- Transit authority hardware validation on a multi-rail GPU node finds a missing NIC. What should operators do?
- A school-district AI lab performs post-install hardware checks on new GPU servers. Which link validation belongs in that gate?
- A housing-authority analytics team reviews day-0 health status on a new GPU server. What indicates the node should not be accepted yet?
- City cabling crews build the GPU fabric for a municipal AI row. How should they choose interconnect media?
- County health-cluster switches are ready for InfiniBand or Ethernet GPU links. What transceiver step belongs in bring-up validation?
- Transit operations finishes pulling GPU fabric cables for a municipal AI factory. What documentation check validates cable types?
- A water-utility AI fabric links with mismatched optics that come up briefly but fail expected signaling margins. What is the correct bring-up conclusion?
- A library consortium AI rack must keep compute and storage cables distinct during bring-up. Which cabling discipline helps?
- A courts OCR cluster designs short-reach GPU links versus longer runs in the same municipal row. When do different cable or transceiver classes apply?
- A city field technician seats a replacement GPU in a municipal HGX server during bring-up. Which physical install practice is required?
- County health racks install additional GPU boards on an AI server. What must be completed before electrical bring-up?
- Transit authority adds GPU capacity and must confirm the physical install before returning a node to service. What integrity check is appropriate?
- A school-district AI lab installs new GPUs and must track physical identity against the purchase order. What install practice is appropriate?
- Emergency-management compute finishes seating GPUs on a standby inference host. When should the chassis be powered on?
- A city permitting cluster must prove new GPUs operate under load before production. Which validation approach fits?
- County public-health nodes show GPUs in idle inventory. What additional gate is required before accepting PHI inference?
- A transit authority runs a municipal train job on new GPUs and wants proof the hardware behaves as expected. What should operators compare?
- Housing-authority analytics sees a node throttle immediately under a known-good test job. What is the correct bring-up response?
- A library consortium completes day-0 workload validation on NLP GPUs. What documentation practice supports later regression compares?
- A city AI factory must reach municipal NFS or parallel storage for training data during bring-up. Which storage step belongs here?
- County health needs datasets reachable on new GPU nodes before BCM imaging finishes. How should storage initial config be treated?
- Transit ops must confirm GPU nodes can use the shared training volume on third-party storage. What should be verified?
- A water-utility digital twin points new GPU servers at storage for checkpoint writes. What endpoint choice is correct at initial config?
- A courts OCR farm configures third-party storage for new GPU servers. Which initial parameters should follow the vendor guide?
Physical Layer Management · 10 questions
- A city enables BlueField on municipal AI nodes to support secure multi-tenant permitting workloads. What is the infrastructure goal?
- County public-health must update BlueField firmware and OS on AI cluster DPUs. Which approach is appropriate?
- Transit authority must set BlueField operation for GPU servers in the municipal AI row. What should guide the mode choice?
- A school-district AI lab finishes BlueField bring-up and must confirm host and DPU software readiness. What validation fits?
- Emergency management must recover a BlueField DPU while the host OS is being rebuilt. What management approach is correct?
- A city must run permitting OCR and 311 chat inference on the same large GPU server safely. Which physical-layer feature applies?
- County health wants HPC-style batch jobs isolated from interactive inference on the same GPU. What should be enabled?
- Transit ops enables MIG mode on municipal GPUs. What post-configuration check is required?
- A housing authority must choose MIG profiles for different civic models on shared GPUs. What should drive the profile selection?
- A library NLP cluster must run a full-GPU training job that cannot use partitions. What MIG action is appropriate?
Control Plane Installation and Configuration · 38 questions
- A city is standing up the control plane for a municipal AI factory and plans HA head nodes. Which product should be installed as the cluster manager?
- County public-health runs a BCM head pair during flu-season inference peaks. What HA outcome must be configured and verified?
- Transit authority installs a secondary BCM head and must confirm HA before declaring the control plane ready. What should operators do?
- A school-district AI lab finishes base BCM install and must onboard the first GPU node group. What should operators do next?
- Emergency management must declare the control plane ready after BCM HA install. What acceptance evidence should be recorded?
- Courts OCR cluster runs a controlled BCM head recovery drill. What must operators confirm?
- A city permitting cluster needs supported OS images on new GPU nodes. How should the OS be installed?
- County public-health selects the host OS for BlueField-ready inference nodes. What OS choice is required?
- Transit ops must rebuild a corrupted GPU node OS while keeping shared fabric settings intact for the rack. What approach is appropriate?
- A water-utility digital-twin rack finishes OS install on GPU nodes. What should operators validate before installing NVIDIA drivers?
- Housing-authority analytics nodes complete OS provisioning from BCM. What confirms install success?
- A city AI factory must differentiate training and inference GPU nodes under BCM. What should administrators configure?
- County health configures BCM networking on GPU nodes for a flu-season model farm. What interface outcome is required?
- Transit authority needs containerized civic training jobs on the new GPU cluster. Which scheduler stack should be installed as part of cluster setup?
- A school-district lab adds new GPU servers and must push consistent software images. When should nodes be assigned to a BCM category?
- Emergency management finishes Slurm and cluster software install on inference nodes. What validation is required?
- A library consortium configures BCM interfaces for GPU nodes running NLP jobs. How should OOB and compute networking be treated?
- City permitting racks finish OS imaging and need GPU software. What driver action is required?
- County health operates BlueField-enabled inference nodes and is updating host GPU software. What additional driver work is required?
- Transit ops must move GPU nodes to a supported NVIDIA driver branch. What should they do with stale drivers first?
- A water utility applies a municipal OS security patch that changes the kernel on GPU nodes. What driver follow-up is required?
- Courts OCR farm rolls a GPU driver update across nodes. How should success be validated?
- Housing authority runs mixed GPU and BlueField hosts and must keep platform software aligned. What versioning practice is required?
- A city needs Docker or Podman on BCM-managed nodes to access GPUs for civic models. What must be installed?
- County health installs NVIDIA Container Toolkit and must enable GPU containers. What configuration step is required?
- Transit authority finishes NVIDIA Container Toolkit deployment on training nodes. How should install success be verified?
- A school-district lab upgrades its container runtime on GPU workstations. What should happen to NVIDIA Container Toolkit?
- Emergency management prepares to go live on an inference category. What Container Toolkit coverage is required first?
- A city permitting desk smoke-tests inference images on GPU hosts with Docker. How should the container access NVIDIA GPUs?
- County public-health starts a Docker container after toolkit install, but nvidia-smi inside the container fails. What should operators check first?
- Transit ops documents required Docker options for municipal AI images that need GPUs. Which guidance belongs in the runbook?
- A library NLP team must confirm CUDA is available inside a Docker container on a shared GPU node. What verification is appropriate?
- Housing analytics prefers a quick GPU sanity check before submitting Slurm or Enroot jobs. What approach fits Domain 3 practice?
- A city needs sanctioned model and container assets from NGC for municipal AI. Where should NGC CLI be installed first?
- County health must pull NGC containers under a municipal organization account. What should operators configure before the first pull?
- Transit authority must place a verification container image onto GPU nodes from NGC. What is the correct NGC CLI use?
- A school-district AI lab policy allows NGC CLI only on approved jump and login hosts. What placement is correct?
- Courts OCR operators installed NGC CLI behind municipal proxies and must prove the tool works. What validation is required?
Cluster Test and Verification · 66 questions
- A city must decide when a new permitting GPU host may join the shared queue. What Domain 4 gate should run first?
- County health sees thermal and power warnings only while single-node GPU stress is running. What should operators conclude?
- Transit ops images many new GPU nodes and needs comparable single-node stress results. What should the standard require?
- A school-district lab runs single-node stress and observes ECC errors and XID storms. What is the correct acceptance action?
- Emergency management must stress newly imaged GPU nodes without jeopardizing 911 NLP standby capacity. How should stress be planned?
- After bring-up, a city wants a floating-point performance baseline on municipal GPU nodes. Which Domain 4 action fits?
- County health finishes HPL on HGX-class inference nodes. How should operators interpret the GFLOPS results?
- Transit ops applies GPU and system firmware updates on a training rack. What HPL practice detects regressions?
- A water-utility digital-twin team needs HPL results comparable across racks. What must be controlled?
- Housing authority acceptance finds one GPU node with an HPL score far below peer nodes. What should operators do?
- A city multi-GPU permitting server must prove intra-node NVLink paths before fabric tests. What should operators run?
- County health single-node NCCL shows asymmetric bandwidth across GPU pairs on an NVLink Switch system. What should operators suspect?
- Transit ops builds a single-node NCCL acceptance suite for new multi-GPU servers. What should the suite include?
- A school-district lab runs single-node NCCL all-reduce. What participation result is required for acceptance?
- Courts OCR finishes hardware install on multi-GPU servers and needs a reference for later regressions. What single-node NCCL practice is required?
- A city must declare an AI fabric row ready and needs proof interconnects are healthy. What cable validation should operators perform?
- County health verification shows BER and signal metrics failing thresholds on one AI fabric cable. What should operators do?
- Transit ops recabled GPU fabric ports and must prove signal health again. What confirmation method fits Domain 4?
- A water-utility AI link shows poor signal quality. Operators must choose the right triage path. What distinction matters?
- Library AI rack validation finishes cable signal checks. What documentation practice is required?
- A city must prove rail-optimized GPU fabric cabling matches design before load. What confirmation is required?
- County health sees links up but NCCL performance far below design after a re-cable. What cabling fault should operators suspect?
- Transit ops must prove each AI fabric cable lands on the intended ports. What verification practice fits?
- Emergency management finds a storage cable seated in a compute rail port on the AI fabric. What should operators do?
- A city prepares to load an AI fabric and must check leaf and spine software currency. What Domain 4 confirmation is required?
- County health finds a leaf switch still on an outdated image during AI fabric validation. What gate action is correct?
- Transit ops completes maintenance on the GPU fabric and must prove switch software consistency. What should operators do?
- A school-district lab upgrades a fabric switch and must close the change. What validation step is required?
- A city must accept DPU-equipped municipal GPU nodes. What BlueField-3 confirmation belongs in Domain 4 verification?
- County health discovery finds a BlueField-3 still on factory firmware during rack acceptance. What should operators do?
- Transit ops defines the contents of every GPU rack acceptance package. What BlueField-3 practice should be included?
- A water utility needs BlueField-3 offload features that depend on DOCA-aligned software. What verification is required?
- A city prepares long-reach AI optics and must validate pluggable modules. What Domain 4 confirmation applies where supported?
- County health finds optics whose firmware is below the approved AI fabric level. What remediation is correct?
- Transit ops performs cable and optics validation walks on the GPU fabric. What transceiver practice should be included?
- Emergency management sees intermittent AI fabric link flaps during verification. What transceiver-related investigation fits Domain 4?
- A city brings up new AI racks and needs a broad node health assessment. What Domain 4 tool should operators run?
- County health ClusterKit output fails in a way that points to network versus GPU. What should operators do with the results?
- Transit ops finishes a major firmware window on the GPU fleet. What ClusterKit practice should follow?
- A school-district lab accepts sibling GPU nodes and wants to spot outliers. How should ClusterKit reports be used?
- Courts OCR completes ClusterKit on new GPU nodes and must keep municipal change evidence. What practice is required?
- A city must prove east-west fabric bandwidth across its municipal AI factory before training jobs scale out. What should operators run?
- County health measures NCCL east-west bandwidth well below the fabric design target. What should operators do next?
- Transit ops adds a new GPU rack and needs a reference for later fabric regressions. What NCCL practice is required?
- A water utility must confirm NCCL performance matches its rail-optimized AI topology. What should verification show?
- Housing authority will not start production training until fabric collectives meet policy. What go-live criterion fits Domain 4?
- A city wants endurance proof on a new permitting rack before tenant jobs arrive. What Domain 4 activity is required?
- County health NCCL burn-in accumulates XID and link errors past policy thresholds. What should operators do?
- Transit ops debates whether a brief NCCL all-reduce equals full acceptance. How should the contrast be framed?
- Emergency management must burn-in fabric nodes without starving 911 NLP inference. What scheduling approach fits?
- Library NLP runs multi-hour NCCL burn-in and needs early warning of latent GPU faults. What should operators capture?
- A city needs proof that municipal GPUs stay stable under sustained Linpack load. What Domain 4 activity is required?
- County health upgrades cooling and must show the change helped under load. What validation fits?
- Transit ops moves GPU nodes from staging into production queues. What HPL requirement should gate that move?
- Water utility HPL burn-in shows progressive score decay over successive hours. What does that signal?
- Courts OCR must schedule multi-hour HPL burn-in without surprising other change windows. What operational control is required?
- A city wants burn-in that stresses AI-framework paths, not only microbenchmarks. What Domain 4 activity fits?
- County health passed HPL and NCCL yet still sees training-stack faults. Why add NeMo burn-in?
- Transit ops configures NeMo burn-in for HGX-class nodes with finite GPU memory. What sizing rule applies?
- School-district lab sees OOM and faults only during NeMo burn-in after HPL/NCCL passed. What should operators do?
- Housing authority prepares an acceptance package after cluster verification. How should NeMo burn-in evidence be handled?
- A city must validate the municipal training filesystem from GPU nodes before large jobs. What Domain 4 step is required?
- County health storage tests report throughput and latency for the dataset pipeline. How should operators interpret them?
- Transit ops validates AI training storage and must cover resiliency, not only peak IOPS. What should storage tests include?
- Water utility storage tests show capacity and paths that look like local disk only, not the shared training export. What should operators conclude?
- A library consortium plans large corpus ingestion onto its municipal GPU training filesystem. What storage verification step should they complete first?
Troubleshoot and Optimize · 24 questions
- A city permitting NOC sees rising GPU temperature and fan alerts on one inference node. What should operators do first?
- County health training jobs show NCCL failures that localize to one host's fabric path. What hardware-fault investigation is most appropriate?
- Transit ops must decide whether a GPU fault on a scheduling node is software-induced or a hardware FRU issue. Which approach fits?
- Emergency-management inference pods die intermittently with PCIe error storms under load. What troubleshooting focus is required?
- A school-district AI lab sees recurring GPU XID errors during exams-week inference. What should admins do with those codes?
- A city multi-GPU municipal server reports ECC failures but does not name which device is bad. What identification step is required?
- County health sees one PSU drop voltage only when AI training peaks, while sibling supplies stay healthy. What conclusion follows?
- Transit fabric errors localize to one PCIe slot hosting a ConnectX adapter on a GPU node. What FRU identification is correct?
- A water-utility AI rack suspects a card fault before opening an RMA. What confirmation step is required?
- Courts OCR ops declare a GPU faulty and must start the RMA path. What linking step is required?
- A city technician must replace a failed GPU in a municipal HGX host. Which replacement practice is required?
- County health must replace a failed PSU on a redundant GPU server. What replacement approach is correct?
- Transit ops replaces a failed NIC on a GPU training node. What must happen before returning the node to the pool?
- A housing authority installs a replacement GPU after a failure. What closeout steps are required?
- A library AI rack must replace dual PSUs over a maintenance window. What ordering rule protects continuity?
- A city wants less training jitter on Intel-based municipal GPU servers. Which optimization path fits Domain 5?
- County health GPU hosts run on AMD servers with uneven training throughput. What optimization guidance applies?
- Transit ops tunes CPU/platform settings across mixed AMD and Intel GPU racks. How should they prove the change helped?
- A school-district lab hears forum advice to overclock municipal GPU servers for 'free performance.' What should they do?
- Emergency management needs lower inference latency on Intel head nodes that currently favor aggressive power saving. What trade-off should they apply?
- City GPU jobs stall on I/O after storage tests show the parallel filesystem is the bottleneck. What optimization fits?
- County health public-health GPUs contend when checkpoints and datasets share one volume. What storage-layout optimization helps?
- Transit ops wants higher storage throughput on GPU nodes after confirming municipal storage can safely accept more concurrency. What tuning is appropriate?
- A water-utility digital-twin team profiles long I/O wait on GPU jobs reading hot feature stores. What optimization decision follows?
These questions are original practice material and are NOT actual exam questions or brain-dump content. All vendor marks are trademarks of their respective owners. This site is not affiliated with, endorsed by, or sponsored by the exam vendor.