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 41

Question 400

During a GPU datacenter maintenance window, an engineer discovers: A 400-Gbps breakout assembly is connected with reversed lane mapping so one logical child port terminates at the wrong peer while other child ports work. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Validate end-to-end cable endpoint identities and intended rail separation before the first collective burn-in.
  2. Require as-built connectivity capture and unique patch labeling as a formal rack handover gate.
  3. Use breakouts' lane identity tests and topology probes in the structured cabling acceptance procedure.
  4. Require application-level NeMo burn-in after changes to scheduler, storage access, and container policy.

Explanation

Correct Answer

C:

This preventive practice addresses the underlying failure (Physical breakout assemblies can expose partial link success while individual logical lanes are miswired.) and supports continued assurance through: Every expected child link resolves to its intended peer and a port-by-port test detects no unintended cross-connection.

Incorrect Answers

A:

This prevention control targets another operational risk: A technician lands two adjacent GPU host fibers on the wrong switch ports, and both links show up yet a multi-rail collective runs over one rail. The currently reported mechanism remains: Physical breakout assemblies can expose partial link success while individual logical lanes are miswired.

B:

This prevention control targets another operational risk: A rack expansion replicates old patch labels, and engineers cannot tell which high-speed trunk belongs to each new GPU server during a fault. The currently reported mechanism remains: Physical breakout assemblies can expose partial link success while individual logical lanes are miswired.

D:

This prevention control targets another operational risk: An LLM platform signs off infrastructure after HPL and NCCL pass, but the real NeMo run fails because the configured container cannot access the required tokenizer and approved model artifacts. The currently reported mechanism remains: Physical breakout assemblies can expose partial link success while individual logical lanes are miswired.

Question 401

A GPU cluster commissioning engineer investigates this situation: GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings. Which corrective deployment or operational action best addresses the stated problem?

  1. Stage compatible fabric software upgrades using verified redundant paths and canary checks according to the network vendor's supported rolling procedure.
  2. Verify consistent supported loss and congestion-control policy across the complete affected fabric, not only device reachability.
  3. Audit switch firmware and software compatibility with port speed, optics, and fabric configuration before applying the supported release.
  4. Collect time-synchronized HPL rank failures, BMC hardware logs, GPU ECC, and node thermal and power readings before replacement.

Explanation

Correct Answer

C:

For the stated NVIDIA AI infrastructure condition (GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings.), this action addresses the concrete failure: Audit switch firmware and software compatibility with port speed, optics, and fabric configuration before applying the supported release. The underlying mechanism is: Switch software versions can alter fabric behavior independently of host drivers and accelerator state. Verification would require: A representative node-to-node test passes with the intended switch firmware, network OS, and port counters recorded.

Incorrect Answers

A:

This option could be appropriate for a different NVIDIA infrastructure problem (4.6: Confirm switch firmware and software): Network engineers push a switch OS update to all spine switches simultaneously, briefly removing every path for long-running distributed training jobs. The present issue is more specifically caused by: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

B:

This option could be appropriate for a different NVIDIA infrastructure problem (4.6: Confirm switch firmware and software): A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices. The present issue is more specifically caused by: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

D:

This option could be appropriate for a different NVIDIA infrastructure problem (4.12: Perform HPL burn-in): A long HPL run fails on the same host every morning, but the report contains no correlated BMC alerts or memory metrics to explain the pattern. The present issue is more specifically caused by: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

Question 402

A GPU cluster commissioning engineer investigates this situation: GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings. Which measured result most directly proves that this specific problem is resolved?

  1. The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic.
  2. A controlled upgrade of one spine at a time maintains protected peer connectivity and the workload meets agreed impairment tolerance.
  3. Restoring the supported firmware combination or link setting returns peer bandwidth with local compute baseline unchanged.
  4. A representative node-to-node test passes with the intended switch firmware, network OS, and port counters recorded.

Explanation

Correct Answer

D:

This is the measurable acceptance result for the described case: A representative node-to-node test passes with the intended switch firmware, network OS, and port counters recorded. It validates the corrective operation (Audit switch firmware and software compatibility with port speed, optics, and fabric configuration before applying the supported release.) rather than a related but different system check.

Incorrect Answers

A:

This observation would help verify a different incident: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. It does not directly establish the required result for the current case: A representative node-to-node test passes with the intended switch firmware, network OS, and port counters recorded.

B:

This observation would help verify a different incident: Network engineers push a switch OS update to all spine switches simultaneously, briefly removing every path for long-running distributed training jobs. It does not directly establish the required result for the current case: A representative node-to-node test passes with the intended switch firmware, network OS, and port counters recorded.

C:

This observation would help verify a different incident: NCCL performance on one server drops after a firmware update while individual GPU memory and compute tests remain steady. It does not directly establish the required result for the current case: A representative node-to-node test passes with the intended switch firmware, network OS, and port counters recorded.

Question 403

A GPU cluster commissioning engineer investigates this situation: GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings. Which root cause or infrastructure principle best explains the observed failure?

  1. Switch software versions can alter fabric behavior independently of host drivers and accelerator state.
  2. Simultaneous changes can turn redundant fabric nodes into one correlated maintenance failure.
  3. Software and platform-state differences between nominally redundant switches can undermine fabric resilience.
  4. A local compute test does not exercise every peer-to-peer accelerator interconnect required by collective operations.

Explanation

Correct Answer

A:

This technical mechanism explains the reported condition: Switch software versions can alter fabric behavior independently of host drivers and accelerator state. The observed situation is: GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings. A requirement-aligned correction is: Audit switch firmware and software compatibility with port speed, optics, and fabric configuration before applying the supported release.

Incorrect Answers

B:

This underlying cause is associated with another GPU platform incident: Network engineers push a switch OS update to all spine switches simultaneously, briefly removing every path for long-running distributed training jobs. The observed case instead concerns: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

C:

This underlying cause is associated with another GPU platform incident: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. The observed case instead concerns: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

D:

This underlying cause is associated with another GPU platform incident: After a GPU tray replacement, CUDA kernels run normally but NCCL tests between certain device pairs fail with fabric or peer-access errors. The observed case instead concerns: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

Question 404

A GPU cluster commissioning engineer investigates this situation: GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Perform software baseline and hardware health consistency checks before accepting redundant switch pairs.
  2. Maintain switch release compatibility and pre/post-upgrade fabric verification in cluster change management.
  3. Version and validate switch congestion policy hop-by-hop whenever fabric software changes.
  4. Bind HPL acceptance to explicit host lists, GPU inventory, workload parameters, and expected scaling efficiency.

Explanation

Correct Answer

B:

This preventive practice addresses the underlying failure (Switch software versions can alter fabric behavior independently of host drivers and accelerator state.) and supports continued assurance through: A representative node-to-node test passes with the intended switch firmware, network OS, and port counters recorded.

Incorrect Answers

A:

This prevention control targets another operational risk: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. The currently reported mechanism remains: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

C:

This prevention control targets another operational risk: A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices. The currently reported mechanism remains: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

D:

This prevention control targets another operational risk: An AI factory reports record HPL throughput but used only half its installed nodes and presents the result as proof of full-cluster performance. The currently reported mechanism remains: Switch software versions can alter fabric behavior independently of host drivers and accelerator state.

Question 405

During NVIDIA AI factory acceptance, a technician observes: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. Which corrective deployment or operational action best addresses the stated problem?

  1. Verify consistent supported loss and congestion-control policy across the complete affected fabric, not only device reachability.
  2. Compare firmware, bootloader, power and event logs with the stable peer and investigate the vendor-documented regression or platform fault.
  3. Audit switch firmware and software compatibility with port speed, optics, and fabric configuration before applying the supported release.
  4. Investigate and retain the failing rank logs, repair its cause, and rerun the intended complete node population.

Explanation

Correct Answer

B:

For the stated NVIDIA AI infrastructure condition (One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision.), this action addresses the concrete failure: Compare firmware, bootloader, power and event logs with the stable peer and investigate the vendor-documented regression or platform fault. The underlying mechanism is: Software and platform-state differences between nominally redundant switches can undermine fabric resilience. Verification would require: The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic.

Incorrect Answers

A:

This option could be appropriate for a different NVIDIA infrastructure problem (4.6: Confirm switch firmware and software): A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices. The present issue is more specifically caused by: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

C:

This option could be appropriate for a different NVIDIA infrastructure problem (4.6: Confirm switch firmware and software): GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings. The present issue is more specifically caused by: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

D:

This option could be appropriate for a different NVIDIA infrastructure problem (4.11: Perform NCCL burn-in): A night-long NCCL burn-in crashes on one rank, yet the operator reruns only successful nodes and presents the new partial result as full-cluster acceptance. The present issue is more specifically caused by: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

Question 406

During NVIDIA AI factory acceptance, a technician observes: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. Which measured result most directly proves that this specific problem is resolved?

  1. A multi-hop transport test carries the expected workload with acceptable drops, congestion signals, and no uncontrolled head-of-line blocking.
  2. A representative node-to-node test passes with the intended switch firmware, network OS, and port counters recorded.
  3. A signed change record includes expected and observed versions and the DPU's managed data path survives a subsequent test.
  4. The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic.

Explanation

Correct Answer

D:

This is the measurable acceptance result for the described case: The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic. It validates the corrective operation (Compare firmware, bootloader, power and event logs with the stable peer and investigate the vendor-documented regression or platform fault.) rather than a related but different system check.

Incorrect Answers

A:

This observation would help verify a different incident: A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices. It does not directly establish the required result for the current case: The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic.

B:

This observation would help verify a different incident: GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings. It does not directly establish the required result for the current case: The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic.

C:

This observation would help verify a different incident: A cluster administrator upgrades BlueField-3 firmware from the host and cannot tell whether the embedded management image committed successfully after reboot. It does not directly establish the required result for the current case: The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic.

Question 407

During NVIDIA AI factory acceptance, a technician observes: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. Which root cause or infrastructure principle best explains the observed failure?

  1. Switch software versions can alter fabric behavior independently of host drivers and accelerator state.
  2. Management connectivity alone does not establish end-to-end loss behavior required by a high-performance GPU Ethernet fabric.
  3. An impressive benchmark score is not meaningful acceptance evidence when the tested topology differs from the deployed cluster.
  4. Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

Explanation

Correct Answer

D:

This technical mechanism explains the reported condition: Software and platform-state differences between nominally redundant switches can undermine fabric resilience. The observed situation is: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. A requirement-aligned correction is: Compare firmware, bootloader, power and event logs with the stable peer and investigate the vendor-documented regression or platform fault.

Incorrect Answers

A:

This underlying cause is associated with another GPU platform incident: GPU nodes lose fabric connectivity after a switch firmware update, but the switch shows no errors because engineers did not check its running network OS version against approved settings. The observed case instead concerns: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

B:

This underlying cause is associated with another GPU platform incident: A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices. The observed case instead concerns: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

C:

This underlying cause is associated with another GPU platform incident: An AI factory reports record HPL throughput but used only half its installed nodes and presents the result as proof of full-cluster performance. The observed case instead concerns: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

Question 408

During NVIDIA AI factory acceptance, a technician observes: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Version and validate switch congestion policy hop-by-hop whenever fabric software changes.
  2. Require topology-aware upgrade sequencing and live path health gates for all AI fabric switches.
  3. Establish per-node healthy GPU throughput and thermal baselines before cross-cluster benchmarking.
  4. Perform software baseline and hardware health consistency checks before accepting redundant switch pairs.

Explanation

Correct Answer

D:

This preventive practice addresses the underlying failure (Software and platform-state differences between nominally redundant switches can undermine fabric resilience.) and supports continued assurance through: The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic.

Incorrect Answers

A:

This prevention control targets another operational risk: A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices. The currently reported mechanism remains: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

B:

This prevention control targets another operational risk: Network engineers push a switch OS update to all spine switches simultaneously, briefly removing every path for long-running distributed training jobs. The currently reported mechanism remains: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

C:

This prevention control targets another operational risk: A GPU worker throttles under stress while facility cooling remains within specification, and operators blame the network because distributed iteration time increases. The currently reported mechanism remains: Software and platform-state differences between nominally redundant switches can undermine fabric resilience.

Question 409

An infrastructure administrator receives the following incident: A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices. Which corrective deployment or operational action best addresses the stated problem?

  1. Compare firmware, bootloader, power and event logs with the stable peer and investigate the vendor-documented regression or platform fault.
  2. Verify consistent supported loss and congestion-control policy across the complete affected fabric, not only device reachability.
  3. Stage compatible fabric software upgrades using verified redundant paths and canary checks according to the network vendor's supported rolling procedure.
  4. Reconcile BlueField-3 firmware, embedded software, and host DOCA/driver versions against the supported interoperability baseline.

Explanation

Correct Answer

B:

For the stated NVIDIA AI infrastructure condition (A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices.), this action addresses the concrete failure: Verify consistent supported loss and congestion-control policy across the complete affected fabric, not only device reachability. The underlying mechanism is: Management connectivity alone does not establish end-to-end loss behavior required by a high-performance GPU Ethernet fabric. Verification would require: A multi-hop transport test carries the expected workload with acceptable drops, congestion signals, and no uncontrolled head-of-line blocking.

Incorrect Answers

A:

This option could be appropriate for a different NVIDIA infrastructure problem (4.6: Confirm switch firmware and software): One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. The present issue is more specifically caused by: Management connectivity alone does not establish end-to-end loss behavior required by a high-performance GPU Ethernet fabric.

C:

This option could be appropriate for a different NVIDIA infrastructure problem (4.6: Confirm switch firmware and software): Network engineers push a switch OS update to all spine switches simultaneously, briefly removing every path for long-running distributed training jobs. The present issue is more specifically caused by: Management connectivity alone does not establish end-to-end loss behavior required by a high-performance GPU Ethernet fabric.

D:

This option could be appropriate for a different NVIDIA infrastructure problem (4.7: Confirm BlueField-3 firmware and software): After BlueField-3 firmware changes, host port links remain up but data path offload tests fail on devices whose DPU-side operating system was not updated. The present issue is more specifically caused by: Management connectivity alone does not establish end-to-end loss behavior required by a high-performance GPU Ethernet fabric.

Question 410

An infrastructure administrator receives the following incident: A rack's Ethernet switches have a functioning management plane but RoCE data links fail because PFC and ECN settings diverge across adjacent devices. Which measured result most directly proves that this specific problem is resolved?

  1. The affected switch completes a sustained stress period on a supported baseline without unexpected reloads or lost cluster traffic.
  2. A controlled upgrade of one spine at a time maintains protected peer connectivity and the workload meets agreed impairment tolerance.
  3. A full GPU-pair coverage test passes after correcting the affected interconnect and records healthy supported fabric links.
  4. A multi-hop transport test carries the expected workload with acceptable drops, congestion signals, and no uncontrolled head-of-line blocking.

Explanation

Correct Answer

D:

This is the measurable acceptance result for the described case: A multi-hop transport test carries the expected workload with acceptable drops, congestion signals, and no uncontrolled head-of-line blocking. It validates the corrective operation (Verify consistent supported loss and congestion-control policy across the complete affected fabric, not only device reachability.) rather than a related but different system check.

Incorrect Answers

A:

This observation would help verify a different incident: One leaf switch reboots repeatedly under load after maintenance, while its peer leaf remains stable and has a different firmware revision. It does not directly establish the required result for the current case: A multi-hop transport test carries the expected workload with acceptable drops, congestion signals, and no uncontrolled head-of-line blocking.

B:

This observation would help verify a different incident: Network engineers push a switch OS update to all spine switches simultaneously, briefly removing every path for long-running distributed training jobs. It does not directly establish the required result for the current case: A multi-hop transport test carries the expected workload with acceptable drops, congestion signals, and no uncontrolled head-of-line blocking.

C:

This observation would help verify a different incident: After a GPU tray replacement, CUDA kernels run normally but NCCL tests between certain device pairs fail with fabric or peer-access errors. It does not directly establish the required result for the current case: A multi-hop transport test carries the expected workload with acceptable drops, congestion signals, and no uncontrolled head-of-line blocking.

NCP-AI Infrastructure (NCP-AII) Practice Tests