NVIDIA AI Infrastructure Professional
Interactive study guides with scenarios, calculators, flashcards, and practice quizzes. Every guide has a "Sample outputs" section showing what its commands print and what to check.
Exam value, from the NCP-AII blueprint (Bring-Up 31%, Physical Layer 5%, Control Plane 19%, Test & Verification 33%, Troubleshooting 12%).
★★★ core: carries heavily weighted objectives (mostly Bring-Up and Test & Verification, plus the BCM core of Control Plane), where scenario questions concentrate.
★★ directly tested objectives with a smaller share of questions.
★ background, or a single niche objective, or beyond the blueprint.
The green numbers are the blueprint objectives each guide covers.
New: filling the gaps NEW
Added to cover the weakest areas against the exam blueprint. Each ends with hands-on simulated tickets.
Command & flag index ★★★NEWall domains401 commands and flags across nvidia-smi, DCGM, NVSM, ipmitool, Redfish, MFT, InfiniBand, nccl-tests, Slurm, Docker, MIG, BCM, and more: what each does and when to use it. Filter by exam domain (e.g. Domains 1 + 4, 64% of the exam) or by objective; searchable table, flip cards with "got it" tracking, and a multiple-choice quiz that follows the filters and remembers your misses
Base Command Manager mastery ★★★NEW3.1 3.2 3.3cmsh and commit, categories vs node overrides, roles and overlay priorities, images and imageupdate, install modes, HA with fencing, Slurm via cm-wlm-setup, MIG through cmsh, lab with five tickets
Cluster burn-in: NCCL, HPL, NeMo ★★★NEW4.11 4.12 4.13Validation vs burn-in, the 1 → N/2 → N sequence, what to monitor, pass criteria, NeMo pretraining in depth (parallelism and throughput math, log reading, checkpoints, failure signatures, a 27-question NeMo set) and the SQuAD question-answering fine-tune, lab with three tickets
GPU drivers, DOCA & Container Toolkit ★★NEW3.4 3.5 3.6Driver + Fabric Manager upgrades, DOCA-OFED replacing MLNX_OFED, nvidia-peermem rebuilds, nvidia-ctk for Docker/containerd, CDI for Podman, error messages, lab with three tickets
Redfish, IPMI & BMC: hands-on lab ★★NEW LAB1.3 5.2Existing guide, now with five tickets: recover a hung node out-of-band, PXE-boot once with raw Redfish (sessions, ETags, Reset), a BMC firmware update via UpdateService/TaskService, fixing an unreachable BMC with cipher suite 0, and collecting RMA evidence for a failed PSU with ipmitool (sdr, sel, fru print, mc info, identify)
DCGM: hands-on lab ★★★NEW LAB4.1 1.10 5.1Existing guide, now with three tickets: find a slow job's real bottleneck with profiling fields instead of GPU_UTIL, confirm an XID 48 with health watches and a targeted level 3 diag, then escalate to vendor support with the evidence, and set up policy, enforced power limits, job stats, and profile pause/resume for Nsight
NVSM: hands-on lab ★★NEW LAB5.1 5.2Existing guide, now with four short tickets: show health and alerts for a failed /raid drive, find a PSU with no input, create the support bundle, and walk the nvsm-> shell
MIG on Kubernetes: GPU Operator MIG Manager ★★NEW LABBeyond the blueprintNew Run:ai section: mig.config labels, single vs mixed strategy, custom layouts, plus a two-ticket kubectl lab: split a node into 1g.10gb, and fix a failed custom layout
HGX firmware & nvfwupd: hands-on lab ★★★NEW LAB1.4Existing guide, now with two tickets: a full DGX H100 update in the documented order (BMC, reset, motherboard tray, GPU tray, ERoT background copy, power cycle, CommitImage), and a multi-system audit that finds staged-but-inactive firmware without killing a running job
Cluster bring-up and validation
AI factory architecture ★★1.1 1.2 1.5BasePOD vs SuperPOD vs Enterprise RAs, scalable units, the four networks, rail-optimized fat tree, switch/cable sizing, rack power, GB200 NVL72, design lab with five tickets
Cluster bring-up & validation ★★★All 39 objectivesPhase-by-phase bring-up, bcm-pod-setup, DGX OS install and upgrade, Slurm/K8s/Run:ai, test matrix
ClusterKit vs other tools ★★4.9ClusterKit, nccl-tests, DCGM: when to use which
HPL mastery ★★★4.2 4.12HPL.dat (and which lines NVIDIA HPL ignores), N/NB/P×Q sizing, Rmax vs Rpeak, every hpl.sh option and HPL_* variable (out-of-core, FP64 emulation), HPL-MxP options, HPCG
gpu-burn & dcgmproftester ★★★4.1 1.10Stress load vs metric validation vs dcgmi diag
GPU, node, and system management
NGC mastery ★★3.3 3.5 3.6 3.7nvcr.io, $oauthtoken login, Enroot/Pyxis, Apptainer, K8s, NGC CLI
CUDA for infrastructure ★★3.4Driver vs runtime, compatibility, env vars, error codes
Error codes ★★5.1 5.2Xid, ECC/row remap, SXid, NVML, CUDA, NCCL, throttle reasons, action ladder, triage drill
nvidia-smi deep dive ★★1.6 2.2Queries, topology, MIG, clocks, persistence
DCGM deep dive ★★★NEW LAB4.1 1.10 5.1Diagnostics, health, policies, fields, exporter, lab with three tickets
Nsight Systems & Compute ★Beyond the blueprintnsys timelines, ncu kernel reports, Slurm profiling, counter permissions, DCGM conflicts
NVSM mastery ★★5.1 5.2DGX health, alerts, dump health, stress-test
Redfish, IPMI & BMC ★★NEW LAB1.3 5.2Out-of-band power/boot, curl & redfishtool, ipmitool, sensors, firmware (push and pull), inventory, accounts, certificates, paging, plus DGX H100/H200 Redfish recipes: resource IDs, HGX_0 and other firmware targets, BIOS settings and boot order, virtual media ISO mounts, GPU power limits, debug bundles; lab with five tickets, 43-question quiz
HGX firmware & nvfwupd ★★★NEW LAB1.4BMC vs HMC, ERoT, .fwpkg, nvfwupd commands, staging vs activation, fleet rollout, plan builder, lab with two tickets
AMD vs Intel server tuning ★★5.4NPS vs SNC, xGMI vs UPI, BIOS power knobs, IOMMU/ACS, OS tuning, NUMA pinning, locality checker
GPU ordering & visibility ★5.4CUDA_VISIBLE_DEVICES, PCI bus order, containers
DGX H100 specs ★1.7GPUs, NVSwitch, NICs, storage, power
Fabric and networking
NVLink & nvbandwidth ★★4.3Link math, NVSwitch topology, bandwidth matrices
Fabric Manager ★★4.3 5.1NVSwitch fabric, modes, partitions, error 802
InfiniBand diagnostics ★★★4.4 4.5 4.10sminfo, smpquery, ibstat, master SM, 400G cables, BER/FEC, mlxlink, lab with six tickets including a permanent master SM move
Networking firmware (MFT) ★★★1.4 4.6 4.7 2.1mst, flint, mlxfwmanager, mlxconfig, mlxfwreset, PSIDs, BlueField BFB, IB/Ethernet switch upgrades, drift checker, five hands-on simulated terminal tickets
Transceivers & signal quality ★★★1.8 4.4 4.5 4.8OSFP/QSFP, DAC/AOC/optics, PAM4 eye diagrams, BER test time, DDM, module firmware
RoCE & Spectrum-X ★1.2 2.1RoCEv2, GIDs, PFC/ECN/DCQCN, SuperNICs, adaptive routing vs InfiniBand
NCCL deep dive ★★4.3 4.10Collectives, env vars, nccl-tests, busbw
nccl-tests mastery ★★★4.3 4.10 4.11Building with MPI, srun vs mpirun, flags, reading output, NCCL_TESTS_SPLIT per-rail and per-node tests, split calculator, six hands-on simulated terminal tickets
NCCL multi-node GPU0 bottleneck ★4.10Rail alignment and HCA mapping troubleshooting