Free NVIDIA NCP-AII practice questions

10 free NVIDIA NCP-AII practice questions with the correct answer and a full explanation for each, taken from the CertStash pack of 170 questions. Work through them, then open each answer to check your reasoning.

Get all 170 questions (US$39) · Download these 10 as a PDF

Question 1

A system engineer needs to set the vGPU scheduling behavior for all GPUs to share the scheduling equally with the default time slice length.

What command should be used?

  1. esxcli system module parameters set -m nvidia -p “NVreg_RegistryDwords=RmPVMRL=0x00”
  2. esxcli system module parameters set -m nvidia -p “NVreg_RegistryDwords=RmPVMRL=0x@1"
  3. esxcli graphics module parameters set -m nvidia -p “NVreg_RegistryDwords=RmPVMRL=0x01”
  4. esxcli system module parameters set -m nvidia -p “NVreg_RegistryDwords=FRL=@x01”
Show answer and explanation

Correct answer: A. esxcli system module parameters set -m nvidia -p “NVreg_RegistryDwords=RmPVMRL=0x00”

“NVreg_RegistryDwords=RmPVMRL=0x00” The correct command uses esxcli system module parameters set with the nvidia module and sets RmPVMRL=0x00 to enable equal GPU scheduling with default time slices. Option A has the correct syntax, module target, and parameter value for this configuration.

Why the other options are wrong

  • B. Uses invalid hex notation @1 instead of proper 0x01 format
  • C. Incorrectly targets graphics module instead of system module
  • D. Uses malformed parameter name FRL with invalid hex notation @x01

Question 2

During a multi-day NeMo burn-in, intermittent “GPU fell off bus” errors occur.

Which diagnostic approach isolates hardware faults?

  1. Run DCGM diagnostics alongside burn-in to monitor GPU health metrics
  2. Switch from BERT to GPT models for simpler computations
  3. Enable HPL_USE_NVSHMEM for alternative memory sharing
  4. Reduce blocksize to 500MB to lower memory pressure
Show answer and explanation

Correct answer: A. Run DCGM diagnostics alongside burn-in to monitor GPU health metrics

GPU health metrics Running DCGM diagnostics alongside burn-in testing provides comprehensive GPU health monitoring and can isolate hardware faults from software issues by tracking thermal, power, and reliability metrics in real-time during stress testing.

Why the other options are wrong

  • B. Changing model type does not isolate hardware faults; it merely shifts the workload
  • C. Enabling NVSHMEM is a memory optimization, not a diagnostic approach
  • D. Reducing blocksize masks symptoms but does not identify underlying hardware problems

Question 3

An engineer needs to verify the current firmware versions of all components (ATF, BSP, NIC, UEFI) on a BlueField-3 DPU’s BMC.

Which Redfish API command provides this information?

  1. mstflint –d <PCI_ID> query full
  2. curl –k –u root:<password> –X GET https://<DPU-BMC-IP>/redfish/v1/UpdateService/FirmwareList
  3. curl –k –u root:<password> –X GET https://<DPU-BMC-IP>/redfish/v1/UpdateService/FirmwareInventory
  4. mlxconfig –d <dev> q
Show answer and explanation

Correct answer: C. curl –k –u root:<password> –X GET https://<DPU-BMC-IP>/redfish/v1/UpdateService/FirmwareInventory

BMC-IP>/redfish/v1/UpdateService/FirmwareInventory The Redfish API endpoint /redfish/v1/UpdateService/FirmwareInventory is the standard interface for querying all firmware component versions on a BlueField-3 DPU's BMC, including ATF, BSP, NIC, and UEFI.

Why the other options are wrong

  • A. mstflint queries firmware via Mellanox tools, not the BMC Redfish API
  • B. FirmwareList is not a valid Redfish endpoint for firmware inventory queries
  • D. mlxconfig queries device configuration parameters, not BMC firmware versions

Question 4

An engineer needs to verify NVLink isolation on a single node with 8 GPUs.

Which NCCL test configuration stresses switch bisection bandwidth?

  1. Use NCCL_TESTS_SPLIT= “DIV 8” with point-to-point tests
  2. Use all_reduce_perf –b 8 –e 16G –f2 –g 8 with NCCL_TESTS_SPLIT= “AND 0x1”
  3. Use all_reduce_perf –b 8 –e 16G –f2 –g 8 without splits
  4. Use reduce_scatter_pref –b 8 –e 16G –f2 –g 4
Show answer and explanation

Correct answer: B. Use all_reduce_perf –b 8 –e 16G –f2 –g 8 with NCCL_TESTS_SPLIT= “AND 0x1”

NCCL_TESTS_SPLIT= “AND 0x1” The all_reduce_perf test with NCCL_TESTS_SPLIT="AND 0x1" stresses switch bisection bandwidth by forcing specific GPU communication patterns that isolate and test the NVLink switch fabric throughput on 8-GPU systems.

Why the other options are wrong

  • A. DIV 8 split is not the correct pattern for bisection bandwidth isolation
  • C. Running without splits does not specifically stress the switch bisection topology
  • D. Using reduce_scatter with only 4 GPUs does not fully exercise the 8-GPU NVLink topology

Question 5

You are validating the environment of an NVIDIA GPU-accelerated data center during post-deployment checks.

Which one action is essential to confirm that power and cooling are sufficient for the stable operation of NVIDIA DGX H100 systems?

  1. Use NVSM to disable unused PCle devices to reduce overall system heat output.
  2. Review the system BIOS to ensure GPU overclocking is enabled for maximum performance.
  3. Verify that each DGX system is connected to redundant, properly rated PDUs and that all power supplies are reporting nominal input.
  4. Confirm the system fans are running at 100% under all workloads to prevent overheating.
Show answer and explanation

Correct answer: C. Verify that each DGX system is connected to redundant, properly rated PDUs and that all power supplies are reporting nominal input.

Verifying redundant power distribution and checking that all PSUs report nominal input voltage is the essential post-deployment check to confirm sufficient and stable power delivery for DGX H100 systems.

Why the other options are wrong

  • A. Disabling PCIe devices reduces power consumption but does not validate that current power is sufficient
  • B. Enabling GPU overclocking increases power demands and heat; it is not a validation practice
  • D. Running fans at 100% continuously is wasteful and indicates inadequate cooling design if needed permanently

Question 6

An engineer needs to validate 400G DAC cable signal integrity in a DGX cluster.

Which CVT metric best identifies marginal cables needing replacement?

  1. Effective BER > 1.5E-254 during a ≤6-hour monitoring window.
  2. Transceiver model matching QSFP-DD specifications.
  3. Temperature fluctuations > 5°C during validation.
  4. Lane power variance < 3dB across all transceivers.
Show answer and explanation

Correct answer: A. Effective BER > 1.5E-254 during a ≤6-hour monitoring window.

An Effective BER (Bit Error Rate) exceeding 1.5E-25 during a monitoring window indicates marginal signal integrity on 400G DAC cables and is the diagnostic metric used by CVT tools to identify cables requiring replacement.

Why the other options are wrong

  • B. Transceiver model matching is a compatibility check, not a signal integrity metric
  • C. Temperature fluctuations are a thermal concern, not a direct signal integrity indicator
  • D. Lane power variance is a power measurement, not a signal integrity metric like BER

Question 7

A user encounters “permission denied” errors when running GPU-accelerated containers on a Secure Boot-enabled system.

What resolves this?

  1. Disable SELinux to relax unnecessary security policies.
  2. Reinstall Docker without the NVIDIA runtime.
  3. Enroll the MOK and sign NVIDIA kernel modules.
  4. Run Docker with sudo for elevated privileges.
Show answer and explanation

Correct answer: C. Enroll the MOK and sign NVIDIA kernel modules.

On Secure Boot-enabled systems, NVIDIA kernel modules must be signed with a Machine Owner Key (MOK). Enrolling the MOK and signing the modules allows the kernel to load them with Secure Boot active, resolving permission denied errors.

Why the other options are wrong

  • A. Disabling SELinux is a security downgrade and does not address Secure Boot module signing requirements
  • B. Reinstalling Docker without the NVIDIA runtime removes GPU support entirely
  • D. Running with sudo bypasses the underlying module loading issue but does not resolve the root cause

Question 8

An InfiniBand administrator needs to run performance benchmarks on new devices added to the fabric.

What tool should be used to check the latency?

  1. tcpdump
  2. ib_write_lat
  3. perfmon
  4. ibdiagnet
Show answer and explanation

Correct answer: B. ib_write_lat

is the InfiniBand latency measurement tool used to benchmark latency on newly added fabric devices, providing precise point-to-point latency metrics.

Why the other options are wrong

  • A. tcpdump is a network packet sniffer, not an InfiniBand-specific latency benchmark tool
  • C. perfmon is a Windows performance monitoring tool, not an InfiniBand utility
  • D. ibdiagnet performs fabric diagnostics and topology checks, not latency measurements

Question 9

A cluster administrator is preparing to update the firmware on a DGX H100 system, including the GPU tray (baseboard).

What is the correct sequence of steps to perform a safe and successful firmware upgrade?

  1. Update the BMC and skip the GPU tray and motherboard tray updates if the system appears healthy.
  2. Update the GPU tray first, then the motherboard tray, and reboot the BMC after all updates are complete.
  3. Stop all GPU activity, update and reboot the BMC, update motherboard and tray components, perform a cold reset, and verify completion.
  4. Perform a cold reset, stop all GPU activity, update and reboot the BMC, update motherboard and tray components, and verify completion.
Show answer and explanation

Correct answer: C. Stop all GPU activity, update and reboot the BMC, update motherboard and tray components, perform a cold reset, and verify completion.

The correct firmware update procedure for DGX H100 systems requires stopping all GPU activity first to ensure no workloads interfere with the update process, then updating and rebooting the BMC as the foundational management controller, followed by updating motherboard and tray components, performing a cold reset to activate all firmware changes, and finally verifying completion to confirm all components are running the new firmware versions.

Why the other options are wrong

  • A. Skipping GPU tray and motherboard updates leaves critical components on outdated firmware, creating stability and security risks.
  • B. Updating GPU tray before BMC is incorrect because the BMC must be updated first to properly manage subsequent component updates and maintain system stability.
  • D. Performing a cold reset before stopping GPU activity and updating the BMC is the wrong sequence and can cause data loss or incomplete updates.

Question 10

One of the nodes in a cluster is not running as fast as the others and the system administrator needs to check the status of the GPUs on that system.

What command should be used?

  1. iblinkinfo
  2. lspci | grep NVIDIA
  3. nvidia-smi
  4. nvidia-gpu-status
Show answer and explanation

Correct answer: C. nvidia-smi

The nvidia-smi command is the standard NVIDIA utility for querying GPU status, performance metrics, memory usage, temperature, and overall health on NVIDIA systems. This is the correct tool for a system administrator to diagnose GPU issues and compare performance across nodes.

Why the other options are wrong

  • A. iblinkinfo is used for InfiniBand link status, not GPU diagnostics.
  • B. lspci grep NVIDIA only lists detected PCI devices and does not provide GPU operational status or performance metrics.
  • D. nvidia-gpu-status is not a standard NVIDIA command for GPU status reporting.

That was 10 of 170.

The full NVIDIA NCP-AII pack has all 170 questions, each with the answer, the explanation and why the other options are wrong, plus a questions-only copy for timed runs. US$39, paid once, with free monthly updates and a pass-or-your-money-back guarantee.

Get the full pack