McAfee-Secured Website

Exam Code: NCP-AII

Exam Name: NCP-AI Infrastructure

Certification Provider: NVIDIA

Questions & Answers

100% NVIDIA NCP-AII Guaranteed Success With Testking Training Materials

Get Real NCP-AII Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

Questions & Answers

170 Questions & Answers with Testing Engine

"NCP-AI Infrastructure Exam", also known as NCP-AII exam, is a NVIDIA certification exam.

Pass your tests with the always up-to-date NCP-AII Exam Engine. Your NCP-AII training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

NVIDIA NCP-AII Practice Test 11

Question 100

During a GPU datacenter maintenance window, an engineer discovers: An operations script identifies GPUs by index rather than UUID, and a reboot changes device enumeration so the wrong card is taken out of production. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Maintain an authoritative serial, slot, bus-address, and GPU-UUID inventory for replacement and monitoring.
  2. Test least-privilege GPU observability for every standardized host image and agent upgrade.
  3. Include GPU enumeration and basic device compute checks in every OS image acceptance test.
  4. Include workload communication locality and transport measurements in placement and topology reviews.

Explanation

Correct Answer

A:

This preventive practice addresses the underlying failure (Enumeration indices may not be durable identifiers across reboots or platform topology changes.) and supports continued assurance through: The selected physical slot consistently maps to the intended UUID after a reboot and maintenance task targets only that device.

Incorrect Answers

B:

This prevention control targets another operational risk: A monitoring agent reports no GPU errors because its account cannot access the required driver interfaces, while local `nvidia-smi` output reveals active device faults. The currently reported mechanism remains: Enumeration indices may not be durable identifiers across reboots or platform topology changes.

C:

This prevention control targets another operational risk: After an OS refresh, a GPU node completes a CPU-only burn-in but its accelerators are absent from the `nvidia-smi` inventory output. The currently reported mechanism remains: Enumeration indices may not be durable identifiers across reboots or platform topology changes.

D:

This prevention control targets another operational risk: A campus-to-data-center AI service relies on long-distance model synchronization, but the proposed network topology ignores the latency effects of cross-region collective operations. The currently reported mechanism remains: Enumeration indices may not be durable identifiers across reboots or platform topology changes.

Question 101

A GPU cluster commissioning engineer investigates this situation: A monitoring agent reports no GPU errors because its account cannot access the required driver interfaces, while local `nvidia-smi` output reveals active device faults. Which corrective deployment or operational action best addresses the stated problem?

  1. Inspect PCIe enumeration, driver installation, and hardware status before declaring GPU workload readiness.
  2. Compare physical slot and cable seating with PCIe enumeration and NVIDIA SMI device identifiers before running training.
  3. Verify the monitoring agent's required permissions and API access without granting unnecessarily broad host privileges.
  4. Restore the approved bring-up order: validate hardware inventory and firmware baseline, then install OS, drivers, and workloads.

Explanation

Correct Answer

C:

For the stated NVIDIA AI infrastructure condition (A monitoring agent reports no GPU errors because its account cannot access the required driver interfaces, while local `nvidia-smi` output reveals active device faults.), this action addresses the concrete failure: Verify the monitoring agent's required permissions and API access without granting unnecessarily broad host privileges. The underlying mechanism is: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware. Verification would require: The read-only monitoring identity collects the expected utilization and health counters and cannot change accelerator configuration.

Incorrect Answers

A:

This option could be appropriate for a different NVIDIA infrastructure problem (1.6: Install GPU-based servers and use NVIDIA SMI): After an OS refresh, a GPU node completes a CPU-only burn-in but its accelerators are absent from the `nvidia-smi` inventory output. The present issue is more specifically caused by: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

B:

This option could be appropriate for a different NVIDIA infrastructure problem (1.6: Install GPU-based servers and use NVIDIA SMI): A server inventory says eight GPUs should be installed, but one accelerator intermittently disappears only after the chassis is moved between racks. The present issue is more specifically caused by: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

D:

This option could be appropriate for a different NVIDIA infrastructure problem (1.1: Describe sequence of events for deployment and validation): A hosting team installs accelerator drivers before confirming BIOS and BMC firmware levels, and several nodes fail post-installation hardware checks. The present issue is more specifically caused by: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

Question 102

A GPU cluster commissioning engineer investigates this situation: A monitoring agent reports no GPU errors because its account cannot access the required driver interfaces, while local `nvidia-smi` output reveals active device faults. Which measured result most directly proves that this specific problem is resolved?

  1. The read-only monitoring identity collects the expected utilization and health counters and cannot change accelerator configuration.
  2. The selected physical slot consistently maps to the intended UUID after a reboot and maintenance task targets only that device.
  3. After correcting the affected physical connection, eight distinct GPU UUIDs remain visible across controlled power cycles.
  4. The link maintains stable optical telemetry and low error counts with the rack doors closed during a full-path stress run.

Explanation

Correct Answer

A:

This is the measurable acceptance result for the described case: The read-only monitoring identity collects the expected utilization and health counters and cannot change accelerator configuration. It validates the corrective operation (Verify the monitoring agent's required permissions and API access without granting unnecessarily broad host privileges.) rather than a related but different system check.

Incorrect Answers

B:

This observation would help verify a different incident: An operations script identifies GPUs by index rather than UUID, and a reboot changes device enumeration so the wrong card is taken out of production. It does not directly establish the required result for the current case: The read-only monitoring identity collects the expected utilization and health counters and cannot change accelerator configuration.

C:

This observation would help verify a different incident: A server inventory says eight GPUs should be installed, but one accelerator intermittently disappears only after the chassis is moved between racks. It does not directly establish the required result for the current case: The read-only monitoring identity collects the expected utilization and health counters and cannot change accelerator configuration.

D:

This observation would help verify a different incident: A rack technician routes active optical cables through tight cable trays, and previously healthy fabric links flap when the door is closed. It does not directly establish the required result for the current case: The read-only monitoring identity collects the expected utilization and health counters and cannot change accelerator configuration.

Question 103

A GPU cluster commissioning engineer investigates this situation: A monitoring agent reports no GPU errors because its account cannot access the required driver interfaces, while local `nvidia-smi` output reveals active device faults. Which root cause or infrastructure principle best explains the observed failure?

  1. Enumeration indices may not be durable identifiers across reboots or platform topology changes.
  2. A physically marginal GPU or interconnect can manifest as intermittent logical device inventory changes.
  3. Local GPU success does not demonstrate readiness of the network and software required for distributed collective training.
  4. Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

Explanation

Correct Answer

D:

This technical mechanism explains the reported condition: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware. The observed situation is: A monitoring agent reports no GPU errors because its account cannot access the required driver interfaces, while local `nvidia-smi` output reveals active device faults. A requirement-aligned correction is: Verify the monitoring agent's required permissions and API access without granting unnecessarily broad host privileges.

Incorrect Answers

A:

This underlying cause is associated with another GPU platform incident: An operations script identifies GPUs by index rather than UUID, and a reboot changes device enumeration so the wrong card is taken out of production. The observed case instead concerns: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

B:

This underlying cause is associated with another GPU platform incident: A server inventory says eight GPUs should be installed, but one accelerator intermittently disappears only after the chassis is moved between racks. The observed case instead concerns: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

C:

This underlying cause is associated with another GPU platform incident: A single-node CUDA test passes, yet a distributed job hangs because the NVIDIA collective communication transport cannot communicate across hosts. The observed case instead concerns: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

Question 104

A GPU cluster commissioning engineer investigates this situation: A monitoring agent reports no GPU errors because its account cannot access the required driver interfaces, while local `nvidia-smi` output reveals active device faults. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Maintain an authoritative serial, slot, bus-address, and GPU-UUID inventory for replacement and monitoring.
  2. Verify accelerator seating, slot mapping, and consistent identity after any physical relocation.
  3. Require hardened BMC initialization with credential rotation and network isolation in the rack acceptance checklist.
  4. Test least-privilege GPU observability for every standardized host image and agent upgrade.

Explanation

Correct Answer

D:

This preventive practice addresses the underlying failure (Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.) and supports continued assurance through: The read-only monitoring identity collects the expected utilization and health counters and cannot change accelerator configuration.

Incorrect Answers

A:

This prevention control targets another operational risk: An operations script identifies GPUs by index rather than UUID, and a reboot changes device enumeration so the wrong card is taken out of production. The currently reported mechanism remains: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

B:

This prevention control targets another operational risk: A server inventory says eight GPUs should be installed, but one accelerator intermittently disappears only after the chassis is moved between racks. The currently reported mechanism remains: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

C:

This prevention control targets another operational risk: A newly received DGX-class server has default BMC credentials and its management interface is reachable from the general campus network. The currently reported mechanism remains: Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.

Question 105

During NVIDIA AI factory acceptance, a technician observes: A newly commissioned host reports an expected GPU count but the installed memory size differs from the purchase specification on two nodes. Which corrective deployment or operational action best addresses the stated problem?

  1. Reconcile `nvidia-smi` hardware inventory and platform serial records against the approved bill of materials before acceptance.
  2. Inspect firmware setup and platform PCIe enablement against the approved GPU server configuration before workload enrollment.
  3. Verify negotiated link speed, supported lane configuration, firmware, and error counters against the intended adapter specification.
  4. Configure unique BMC credentials, isolated out-of-band addressing, role-limited access, and an audited first-login process.

Explanation

Correct Answer

A:

For the stated NVIDIA AI infrastructure condition (A newly commissioned host reports an expected GPU count but the installed memory size differs from the purchase specification on two nodes.), this action addresses the concrete failure: Reconcile `nvidia-smi` hardware inventory and platform serial records against the approved bill of materials before acceptance. The underlying mechanism is: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity. Verification would require: An independently sampled inventory records correct GPU model, quantity, memory, firmware, and server identifiers for all accepted hosts.

Incorrect Answers

B:

This option could be appropriate for a different NVIDIA infrastructure problem (1.7: Validate installed hardware): A provisioning script accepts a data center node after disk health checks, but the BIOS is configured to hide required PCIe slots and GPUs are unavailable. The present issue is more specifically caused by: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

C:

This option could be appropriate for a different NVIDIA infrastructure problem (1.7: Validate installed hardware): A technician sees a healthy network interface LED on a ConnectX adapter but a cluster workload receives only a fraction of the planned throughput. The present issue is more specifically caused by: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

D:

This option could be appropriate for a different NVIDIA infrastructure problem (1.3: Perform initial configuration of BMC, out-of-band management, and TPM): A newly received DGX-class server has default BMC credentials and its management interface is reachable from the general campus network. The present issue is more specifically caused by: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

Question 106

During NVIDIA AI factory acceptance, a technician observes: A newly commissioned host reports an expected GPU count but the installed memory size differs from the purchase specification on two nodes. Which measured result most directly proves that this specific problem is resolved?

  1. Each host's power supply inventory matches the approved platform matrix and sustained stress leaves redundancy headroom intact.
  2. PCIe topology and accelerator inventory match the expected slot map and the host passes a GPU-specific smoke test.
  3. The staged bring-up record maps each GPU chassis to dual feeds and the controlled PDU outage leaves protected nodes powered.
  4. An independently sampled inventory records correct GPU model, quantity, memory, firmware, and server identifiers for all accepted hosts.

Explanation

Correct Answer

D:

This is the measurable acceptance result for the described case: An independently sampled inventory records correct GPU model, quantity, memory, firmware, and server identifiers for all accepted hosts. It validates the corrective operation (Reconcile `nvidia-smi` hardware inventory and platform serial records against the approved bill of materials before acceptance.) rather than a related but different system check.

Incorrect Answers

A:

This observation would help verify a different incident: A recently delivered accelerator system has inconsistent PSU models across sibling nodes, and unexpected power alarms occur only during high utilization. It does not directly establish the required result for the current case: An independently sampled inventory records correct GPU model, quantity, memory, firmware, and server identifiers for all accepted hosts.

B:

This observation would help verify a different incident: A provisioning script accepts a data center node after disk health checks, but the BIOS is configured to hide required PCIe slots and GPUs are unavailable. It does not directly establish the required result for the current case: An independently sampled inventory records correct GPU model, quantity, memory, firmware, and server identifiers for all accepted hosts.

C:

This observation would help verify a different incident: A new GPU pod is powered on before rack inventory and PDU mapping are approved, and engineers cannot identify which chassis will lose power during maintenance. It does not directly establish the required result for the current case: An independently sampled inventory records correct GPU model, quantity, memory, firmware, and server identifiers for all accepted hosts.

Question 107

During NVIDIA AI factory acceptance, a technician observes: A newly commissioned host reports an expected GPU count but the installed memory size differs from the purchase specification on two nodes. Which root cause or infrastructure principle best explains the observed failure?

  1. A physical link-light state does not prove full-width and error-free transport capacity.
  2. A valid storage and operating system image cannot compensate for platform firmware that disables accelerator resources.
  3. Missing telemetry can reflect access or software integration gaps rather than healthy accelerator hardware.
  4. Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

Explanation

Correct Answer

D:

This technical mechanism explains the reported condition: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity. The observed situation is: A newly commissioned host reports an expected GPU count but the installed memory size differs from the purchase specification on two nodes. A requirement-aligned correction is: Reconcile `nvidia-smi` hardware inventory and platform serial records against the approved bill of materials before acceptance.

Incorrect Answers

A:

This underlying cause is associated with another GPU platform incident: A technician sees a healthy network interface LED on a ConnectX adapter but a cluster workload receives only a fraction of the planned throughput. The observed case instead concerns: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

B:

This underlying cause is associated with another GPU platform incident: A provisioning script accepts a data center node after disk health checks, but the BIOS is configured to hide required PCIe slots and GPUs are unavailable. The observed case instead concerns: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

C:

This underlying cause is associated with another GPU platform incident: A monitoring agent reports no GPU errors because its account cannot access the required driver interfaces, while local `nvidia-smi` output reveals active device faults. The observed case instead concerns: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

Question 108

During NVIDIA AI factory acceptance, a technician observes: A newly commissioned host reports an expected GPU count but the installed memory size differs from the purchase specification on two nodes. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Automate bill-of-materials reconciliation with actual hardware discovery during rack handover.
  2. Collect link capability and negotiated-state evidence instead of relying on port LEDs at acceptance.
  3. Define ECC threshold monitoring and memory test criteria in the standard hardware validation plan.
  4. Include inter-node collective workload checks before admitting GPU servers to a production training queue.

Explanation

Correct Answer

A:

This preventive practice addresses the underlying failure (Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.) and supports continued assurance through: An independently sampled inventory records correct GPU model, quantity, memory, firmware, and server identifiers for all accepted hosts.

Incorrect Answers

B:

This prevention control targets another operational risk: A technician sees a healthy network interface LED on a ConnectX adapter but a cluster workload receives only a fraction of the planned throughput. The currently reported mechanism remains: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

C:

This prevention control targets another operational risk: A GPU host is labeled fully functional based on a POST screen even though one DIMM records recurrent correctable ECC faults under memory load. The currently reported mechanism remains: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

D:

This prevention control targets another operational risk: A single-node CUDA test passes, yet a distributed job hangs because the NVIDIA collective communication transport cannot communicate across hosts. The currently reported mechanism remains: Counting accelerator devices alone does not verify that the installed configuration matches contracted capacity.

Question 109

An infrastructure administrator receives the following incident: A technician sees a healthy network interface LED on a ConnectX adapter but a cluster workload receives only a fraction of the planned throughput. Which corrective deployment or operational action best addresses the stated problem?

  1. Compare the installed PSU part numbers and supported redundancy capacity with the vendor's approved system configuration.
  2. Reconcile `nvidia-smi` hardware inventory and platform serial records against the approved bill of materials before acceptance.
  3. Verify negotiated link speed, supported lane configuration, firmware, and error counters against the intended adapter specification.
  4. Inspect PCIe enumeration, driver installation, and hardware status before declaring GPU workload readiness.

Explanation

Correct Answer

C:

For the stated NVIDIA AI infrastructure condition (A technician sees a healthy network interface LED on a ConnectX adapter but a cluster workload receives only a fraction of the planned throughput.), this action addresses the concrete failure: Verify negotiated link speed, supported lane configuration, firmware, and error counters against the intended adapter specification. The underlying mechanism is: A physical link-light state does not prove full-width and error-free transport capacity. Verification would require: Link diagnostics report expected width and rate with clean error counters and a controlled throughput test meets the acceptance floor.

Incorrect Answers

A:

This option could be appropriate for a different NVIDIA infrastructure problem (1.7: Validate installed hardware): A recently delivered accelerator system has inconsistent PSU models across sibling nodes, and unexpected power alarms occur only during high utilization. The present issue is more specifically caused by: A physical link-light state does not prove full-width and error-free transport capacity.

B:

This option could be appropriate for a different NVIDIA infrastructure problem (1.7: Validate installed hardware): A newly commissioned host reports an expected GPU count but the installed memory size differs from the purchase specification on two nodes. The present issue is more specifically caused by: A physical link-light state does not prove full-width and error-free transport capacity.

D:

This option could be appropriate for a different NVIDIA infrastructure problem (1.6: Install GPU-based servers and use NVIDIA SMI): After an OS refresh, a GPU node completes a CPU-only burn-in but its accelerators are absent from the `nvidia-smi` inventory output. The present issue is more specifically caused by: A physical link-light state does not prove full-width and error-free transport capacity.

Question 110

An infrastructure administrator receives the following incident: A technician sees a healthy network interface LED on a ConnectX adapter but a cluster workload receives only a fraction of the planned throughput. Which measured result most directly proves that this specific problem is resolved?

  1. Each host's power supply inventory matches the approved platform matrix and sustained stress leaves redundancy headroom intact.
  2. PCIe topology and accelerator inventory match the expected slot map and the host passes a GPU-specific smoke test.
  3. A representative all-reduce run across distant racks reaches the approved throughput and remains operational after one specified uplink failure.
  4. Link diagnostics report expected width and rate with clean error counters and a controlled throughput test meets the acceptance floor.

Explanation

Correct Answer

D:

This is the measurable acceptance result for the described case: Link diagnostics report expected width and rate with clean error counters and a controlled throughput test meets the acceptance floor. It validates the corrective operation (Verify negotiated link speed, supported lane configuration, firmware, and error counters against the intended adapter specification.) rather than a related but different system check.

Incorrect Answers

A:

This observation would help verify a different incident: A recently delivered accelerator system has inconsistent PSU models across sibling nodes, and unexpected power alarms occur only during high utilization. It does not directly establish the required result for the current case: Link diagnostics report expected width and rate with clean error counters and a controlled throughput test meets the acceptance floor.

B:

This observation would help verify a different incident: A provisioning script accepts a data center node after disk health checks, but the BIOS is configured to hide required PCIe slots and GPUs are unavailable. It does not directly establish the required result for the current case: Link diagnostics report expected width and rate with clean error counters and a controlled throughput test meets the acceptance floor.

C:

This observation would help verify a different incident: A training farm connects every GPU node through one oversubscribed top-of-rack switch even though collective jobs exchange large volumes of east-west gradients. It does not directly establish the required result for the current case: Link diagnostics report expected width and rate with clean error counters and a controlled throughput test meets the acceptance floor.

NCP-AI Infrastructure (NCP-AII) Practice Tests