Free NVIDIA-Certified Professional: AI Rack and Interconnect (NCP-ARI) Practice Questions

Sample NCP-ARI Practice Questions

Browse all 20 free NVIDIA-Certified Professional: AI Rack and Interconnect practice questions below.

1. Your greenfield hall will host air-cooled DGX H100 compute racks now and a liquid-cooled GB200 NVL72 row in a later phase. The electrical contractor asks how power distribution inside the two rack types differs so the panel schedules can be planned correctly. Which description is accurate?

   • NVL72 racks keep conventional AC rPDUs for their compute trays and reserve the DC busbar exclusively for the NVLink switch trays mounted in the center of the rack.
   • DGX H100 racks feed three-phase AC through dual rPDUs to each system's internal supplies, while NVL72 power shelves rectify AC and drive the compute trays over a DC busbar.
   • DGX H100 systems accept a 54 V DC feed from centralized rectifier shelves, while each NVL72 compute tray carries its own AC-input power supplies fed from rPDUs.
   • Both rack types hand AC to the equipment, but the NVL72 raises its distribution to 415 V line-to-line so that each compute tray draws fewer amps across its dual input cords.

2. The pathway design for the scalable unit runs inter-row InfiniBand bundles on overhead ladder rack between the compute rows and the spine position. During the site survey you must confirm the planned pathway can actually carry the deployment. Which check correctly qualifies the ladder rack?

   • Compare projected cable weight per unit length, with planned future fills, against the tray's rated load at the support span as installed.
   • Confirm the hangers land every 3 meters to match the ceiling grid, the default spacing that published tray load ratings are quoted against.
   • Count the cables against the tray's cross-sectional fill limit, since the published fill ratios are what ultimately govern allowable tray weight.
   • Match the tray gauge to the heaviest single bundle it must carry, because point loading is what sizes ladder rack rather than distributed weight.

3. You are auditing the dressing practices of a new subcontractor on the first completed rack of the greenfield SuperPOD scalable unit, checking their work against NVIDIA's cabling guidance before releasing them to cable the remaining racks. Which two observed practices should pass the audit? (Select TWO)

   • Bundles fastened with soft hook-and-loop wraps snug enough to hold the group but loose enough to leave every jacket undeformed.
   • Bundles tied off to the rack structure so connector bodies carry no cable weight when doors are closed or the run is disturbed.
   • Thin releasable zip ties used on straight tray sections, with hook-and-loop reserved for bends where contact pressure concentrates.
   • Copper and fiber for one switch combined into a single tight harness, so one wrap spacing and label scheme serves the whole group.
   • Wraps cinched until each bundle holds a rigid round profile, so the harness self-supports and needs fewer anchor points per run.

4. During link bring-up in the new pod, one DR4 uplink shows no link LED while every other link in the same trunk bundle came up clean. Ports are enabled and the module is recognized on both ends. Following the standard fault isolation sequence, what should the technician do first?

   • Inspect and clean both end-faces, re-seat the connectors and the module, and retest the link before swapping any hardware at all.
   • Flip the trunk's polarity at the cassette first, since a Type A versus Type B mismatch produces exactly this single dark link pattern.
   • Move the circuit onto a spare trunk fiber pair first, since a damaged fiber in a fresh pull is the likeliest fault and quick to isolate.
   • Swap the transceiver for a known-good spare first, since module infant mortality outranks contamination on freshly installed links.

5. An rPDU in one of the greenfield pod's high-density AC racks must be replaced. Its input whip lands on an overhead busway tap serving the row, and adjacent racks stay in production during the work. What is the correct energy control sequence before the technician touches the whip?

   • Apply lockout tagout at the tap and begin work directly, since a locked disconnect makes voltage verification a redundant step.
   • Have a qualified person apply lockout tagout at the busway tap, verify absence of voltage at the whip, and then begin the swap.
   • Verify absence of voltage at the whip first, then apply lockout at the tap, keeping the test aligned with the working position.
   • Switch the rPDU's input breaker off and begin the work, since that breaker is the nearest disconnecting means to the equipment.

6. A support rack for an NVL72 pod will draw a continuous 26 kW across dual feeds. The electrical contractor offers 60 A, 415 V three-phase whips, and the technician must state the usable continuous capacity of one whip after applying the NEC 80 percent continuous-load rule. Which figure and derivation are right?

   • About 43.1 kW, from the square root of three times 415 V times the full 60 A breaker rating.
   • About 29.9 kW, from three times the 240 V phase-to-neutral voltage times a 41.5 A limit.
   • About 34.5 kW, from the square root of three times 415 V times the 48 A continuous limit.
   • About 19.9 kW, from 415 V multiplied directly by the 48 A continuous limit on one phase pair.

7. While planning intra-rack routing in a GB200 NVL72 rack, a technician notes the rear center is occupied by the blind-mate NVLink cartridge backplane and the DC busbar. A colleague proposes dropping management cabling straight down the rear center anyway because it shortens every run. What is the right routing plan?

   • Route the cabling outside the rack along the containment frame, since NVL72 racks reserve their internal channels for coolant hoses and busbar service.
   • Use the center path but sleeve the cables in split loom, since a protective sleeve addresses the guidance against routing cable through the middle of a rack.
   • Route cabling in the side channels the extra-wide frame provides, keeping the center clear so the busbar and blind-mate backplane zone stay serviceable.
   • Use the center path for the low-voltage management runs only, reserving the side channels exclusively for the fabric bundles that carry the density load.

8. You are finalizing the trunk order for the NVL72 pod's leaf-to-spine fiber plant, and the procurement sheet must state fiber type, connector polish, and reach class for each optic family before it goes to the vendor. Which two statements should drive the entries on that sheet? (Select TWO)

   • DR4 optics launch into 50 um single-mode fiber, so single-mode and multimode trunks stay interchangeable at 400G speeds.
   • SR4 optics need OM4 multimode and remain inside budget out to roughly 150 m once RS-FEC is enabled on both ends.
   • DR4 optics launch into 9 um single-mode fiber through MPO connectors finished with an 8 degree APC polish.
   • SR4 optics need 50 um OM3 or OM4 multimode trunks and stay in budget only to about 50 m of channel length.
   • DR4 optics use flat-polish MPO/PC ferrules, letting one trunk build serve both PC and APC panel ports interchangeably.

9. A 40-cable inter-rack bundle leaves the overhead pathway and drops toward an NVL72 network rack. The installer has anchored it at the tray exit and at the rack entry, 3.5 m apart, and the span between sags visibly under its own weight. What does NVIDIA's cabling guidance require for this section of the run?

   • Add supports so the bundle is carried at least every 1.5 m along the run, matching the same support interval the overhead pathway itself uses.
   • Add supports so the bundle is carried at least every 60 cm along the run, and guide roughly the first 8 feet into cable management before continuing.
   • Leave the span as dressed, since sag between anchors is acceptable when both ends are strain-relieved and the low point clears the rack door swing.
   • Tension the bundle between the two anchors until the sag disappears, since a straight span loads the anchor straps instead of the cables themselves.

10. After the connect phase on eighteen racks, the integrator needs to confirm that several thousand fabric connections match the physical network design before InfiniBand bring-up, and swapped leaf ports are the defect they most expect from a crew of this size. Which verification approach is purpose-built for this check?

    • Run a full fabric diagnostic sweep once workloads start, since congestion counters expose miscabled links faster than a static topology comparison.
    • Run the NVIDIA Cable Validation Tool against the intended topology, comparing actual port-to-port connectivity with the design to flag miscabling.
    • Send a second crew to re-verify every link visually against the printed cable schedule, since tooling cannot read labels and labels carry design intent.
    • Check that link LEDs come up across the fabric, since a swapped pair of same-speed leaf ports would hold at least one of the links in a down state.

11. A brownfield hall mixes legacy air-cooled DGX racks with a new GB200 NVL72 rack arriving next month. A junior technician asks how power distribution inside the NVL72 rack differs from the rPDU-fed racks already on the floor. Which statement gives the accurate difference the install plan must reflect?

    • Extra-wide rails let the rack host vertically mounted high-amperage rPDUs whose outlet banks feed each tray, matching the A/B cord model of the legacy racks.
    • Power shelves in the rack rectify the facility AC input and energize a DC busbar, so compute trays draw DC from the busbar instead of plugging cords into rPDU outlets.
    • The rack's vertical busbar distributes three-phase AC directly to each compute tray, which removes the need for the outlet-level metering that the legacy racks' rPDUs currently provide.
    • Facility-supplied DC arrives at the rack's power shelves, which then step the voltage down for the busbar, so the electrical room must add rectifier plant first.

12. The expansion introduces XDR leaf switches into a fabric whose production pods run NDR, and the project's cable schedule template comes from the original build. During documentation review you must decide what each schedule row must capture for the new links. Which requirement is the correct one?

    • The original NDR schedule can be cloned since port counts and topology match, with a global note substituting the XDR part families across the whole document.
    • Each row carries bundle identifiers and aggregate counts per rack pair, since per-port detail lives in the switch configuration and would duplicate that data.
    • Each row identifies the switch-side end by rack, U position, and port, with the server end recorded later at install, keeping the schedule stable through design churn.
    • Each row identifies both link ends by rack, U position, and port, plus the specific cable or transceiver part number and length validated for that link's generation.

13. A crew dresses mixed NDR and XDR bundles in the expansion racks while the neighboring pod carries production traffic. The lead reviews fastening and strain practice against the DGX SuperPOD cabling guide before sign-off of the completed racks. Which two practices belong on the sign-off checklist? (Select TWO)

    • Secure the bundles with soft hook-and-loop straps cinched only until snug.
    • Tie the bundles to the rack structure so connector shells carry no load.
    • Anchor the bundles to the adjacent power whips to reuse proven tie points.
    • Fasten with thin releasable zip ties torqued by hand to speed later rework.
    • Snug the straps until jackets flatten slightly, confirming firm retention.

14. Between the tray end and the first expansion rack, a new XDR trunk bundle crosses a 2.4 m gap above the cold-aisle containment with a single midpoint support, leaving unsupported spans of about 1.2 m. The bundle already shows a slight catenary sag at both spans. What does the cabling practice require here?

    • Accept spans up to 1.5 m for fiber trunks, whose low linear mass keeps catenary strain inside jacket limits.
    • Add supports so no span along the run exceeds roughly 60 cm, about every two feet of cable.
    • Support only at direction changes and terminations, letting the straight runs float between anchor points.
    • Sleeve the gap in split loom so the bundle's own stiffness carries the span without any added hardware.

15. A technician must verify phase balance at an energized 415 V three-phase remote power panel serving the expansion row. The panel carries an arc flash label from last year's incident energy study, and the hall's PPE locker stocks several categories of gear. What governs what the technician wears and where he stands?

    • Arc flash controls begin above 600 V, so a 415 V panel calls for shock PPE and insulated tools by themselves.
    • The label's incident energy and arc flash boundary set the PPE category and the standoff for this task.
    • The boundary takes effect once covers come off, so closed-door verification needs standard gloves at most.
    • A category 2, 8 cal per cm2 kit covers data center panels, which fall below the study's analysis threshold.

16. A colocation provider preparing whips for an incoming GB200 NVL72 rack asks whether the tenant needs the facility's negative 48 V DC plant, since the rack is advertised as using DC busbar power distribution to its compute trays. How should the deployment technician explain the rack's actual input requirement?

    • The rack needs the facility DC plant, because the NVLink cartridge backplane is fed straight from the busbar and cannot tolerate rectifier ripple at full load.
    • The rack needs AC only for its fans and management plane, while the compute trays pull their main load from the facility's separate DC distribution plant.
    • The rack accepts either AC whips or the 48 V plant, because its power shelves auto-range between rectifier duty and a direct DC pass-through operating mode.
    • The rack takes standard three-phase AC whips; its internal power shelves rectify that input and energize the DC busbar that serves the compute trays.

17. Before a colocation build, a lead technician reviews the physical network documentation the pull teams will work from. The cluster uses fixed-length LinkX cables ordered ahead of time, and the landlord requires a routing permit for every pathway segment a tenant uses. Which content standard should the cable schedule review enforce?

    • Each row lists both terminations by rack, U, and port, plus the cable part number with length code, the label ID, and the approved pathway route.
    • Each row lists the cable part number, length code, and label ID, with the routing left to the pull team's discretion inside the permitted tray network.
    • Each row lists the switch-end termination and the label ID, with server-end ports assigned by the rack technician at pull time to absorb any field variances.
    • Each row lists both terminations and the pathway route, with lengths recorded as measured after installation so the as-built reflects true pulled distances.

18. A colocation landlord offers free use of standard 600 mm wide cabinets, but the deployment design for a SuperPOD-style build specifies extra-wide racks at the switch positions. The project manager asks a technician to justify paying for the wider footprint out of the project budget. Which justification is technically sound?

    • The added width enlarges the front-to-rear plenum, letting the cabinets carry the same load with fewer perforated panels in the containment doors.
    • The added width creates side channels for vertical managers and rPDUs, letting high-count cable bundles route beside the rails clear of the equipment.
    • The added width increases usable mounting depth, which the longer chassis of AI switch platforms need for rear service access and airflow clearance.
    • The added width allows two half-width switches to mount side by side per U, cutting the rack count the landlord charges against the cage license.

19. A colocation build lead is writing the bundle-dressing section of the method of procedure for NDR copper and MPO fiber runs through the facility's shared trays. The crews are mixed-experience contractors, so each instruction must match DGX SuperPOD cabling guidance exactly. Which two practices belong in the procedure? (Select TWO)

    • Fasten with thin plastic ties at wide spacing, trimmed flush so the cut ends cannot abrade neighboring jackets.
    • Fasten with soft hook-and-loop ties snugged so cable jackets stay round and undeformed under every wrap point.
    • Support each bundle at least every 4 feet along the tray route, halving the hardware while staying inside guidance.
    • Guide roughly the first 8 feet of each bundle into cable management before continuing to pull the remainder.
    • Introduce connector reorientation twists inside the final 30 cm so alignment is established right at the cage face.

20. After landing about 1,500 links across two colocation cages, a build lead needs to validate the cabling at scale before handoff: link state, module identification data, and error counters across the whole fabric, with reporting the customer can archive as acceptance evidence. Which NVIDIA tool fits this stage of the deployment?

    • An optical power meter survey of every link, the method NVIDIA prescribes ahead of any software-based validation pass on the fabric.
    • The UFM telemetry platform, which streams per-port counters and serves as NVIDIA's designated cable acceptance authority for handoff.
    • The mlxlink utility executed per port from each host, which NVIDIA positions as the fabric-wide validation report generator for builds.
    • The Cable Validation Tool (CVT), which checks the fabric's cables and transceivers at scale and produces validation reporting for the build.