cert
cert-1
cert-2

Pass NVIDIA NCP-AII Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers to Pass the Actual Exam!
30 Days Free Updates, Instant Download!

cert-5
cert-6
NCP-AII Exam - Verified By Experts
NCP-AII Premium File

NCP-AII Premium File

$89.99
$98.99
  • Premium File 170 Questions & Answers. Last Update: Nov 06, 2026

Whats Included:

  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates
 
$98.99
$89.99
accept 14 downloads in the last 7 days
block-screenshots
NVIDIA NCP-AII Practice Test 112

Question 1110

During an NVIDIA AI factory acceptance test, engineers discover: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes. Which TWO complementary elements are needed: the correct technical action and the strongest independent diagnostic evidence? Select TWO answers.

  1. Quarantine and diagnose affected GPUs using ECC, row-remap and diagnostic evidence before authorized repair or return.
  2. Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence.
  3. Compare bidirectional port counters, cable swaps, PCIe link state, adapter logs and test transfers.
  4. Inspect lscpu NUMA topology, numactl placement, memory bandwidth, CPU affinity and GPU data-feed stalls.
  5. Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component.

Explanation

Correct Answers

C:

This is the direct operational evidence to investigate: Compare bidirectional port counters, cable swaps, PCIe link state, adapter logs and test transfers. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes.

E:

This is an appropriate correction for Failed ConnectX HCA or port fault isolation: Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes. The governing mechanism is: A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU.

Incorrect Answers

A:

This alternative is relevant to GPU ECC and memory reliability diagnosis but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: Quarantine and diagnose affected GPUs using ECC, row-remap and diagnostic evidence before authorized repair or return. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes.

B:

This tempting but unsafe shortcut is not a sufficient remediation for Failed ConnectX HCA or port fault isolation: Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence. A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes. A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU.

D:

This alternative is relevant to AMD host CPU NUMA and memory tuning but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: Inspect lscpu NUMA topology, numactl placement, memory bandwidth, CPU affinity and GPU data-feed stalls. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes.

Question 1111

A production GPU cluster review identifies this condition: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes. Which diagnostic evidence should be obtained FIRST to substantiate the likely fault or configuration gap? Select ONE best answer.

  1. Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence.
  2. Compare bidirectional port counters, cable swaps, PCIe link state, adapter logs and test transfers.
  3. Inspect nvidia-smi ECC history, DCGM alerts, remapped rows, Xid logs and stress test results.
  4. Inspect fan tachometer and BMC sensors, GPU temperature, throttle reasons, inlet mapping and workload timing.

Explanation

Correct Answer

B:

This is the direct operational evidence to investigate: Compare bidirectional port counters, cable swaps, PCIe link state, adapter logs and test transfers. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes. The governing mechanism is: A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for Failed ConnectX HCA or port fault isolation: Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes. The governing mechanism is: A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU. The proportionate action is: Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component.

C:

This alternative is relevant to GPU ECC and memory reliability diagnosis but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: Inspect nvidia-smi ECC history, DCGM alerts, remapped rows, Xid logs and stress test results. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes.

D:

This alternative is relevant to Fan faults and thermal throttling investigation but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: Inspect fan tachometer and BMC sensors, GPU temperature, throttle reasons, inlet mapping and workload timing. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes.

Question 1112

In a professional AI infrastructure deployment, a technician observes: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes. After corrective work, which observable result BEST confirms that the requirement is met? Select ONE best answer.

  1. Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence.
  2. The repaired path maintains intended negotiated performance with acceptable errors under repeated RDMA stress.
  3. Restored service matches approved acceptance tests and the responsible failure mode has a documented preventative action.
  4. The identified cause is addressed and repeated representative work runs without the original Xid failure.

Explanation

Correct Answer

B:

This is an observable verified result, not merely a claimed configuration change: The repaired path maintains intended negotiated performance with acceptable errors under repeated RDMA stress. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for Failed ConnectX HCA or port fault isolation: Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes. The governing mechanism is: A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU. The proportionate action is: Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component. A direct check is: Compare bidirectional port counters, cable swaps, PCIe link state, adapter logs and test transfers. The acceptance condition is: The repaired path maintains intended negotiated performance with acceptable errors under repeated RDMA stress.

C:

This alternative is relevant to Operational fault isolation and safe rollback but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: Restored service matches approved acceptance tests and the responsible failure mode has a documented preventative action. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes.

D:

This alternative is relevant to NVIDIA GPU Xid event root cause isolation but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: The identified cause is addressed and repeated representative work runs without the original Xid failure. Observed scenario: A replacement high-speed adapter is installed in the wrong slot and negotiates fewer PCIe lanes.

Question 1113

The cluster operations team needs to evaluate this situation: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault. Which operational response is MOST appropriate to address the root issue? Select ONE best answer.

  1. Use a staged evidence-led incident workflow, isolate the nearest failing layer and validate a safe known-good rollback.
  2. Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence.
  3. Validate Intel CPU, PCIe and NUMA placement with supported BIOS and stable sustained-power settings.
  4. Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component.

Explanation

Correct Answer

D:

This is an appropriate correction for Failed ConnectX HCA or port fault isolation: Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

Incorrect Answers

A:

This alternative is relevant to Operational fault isolation and safe rollback but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: Use a staged evidence-led incident workflow, isolate the nearest failing layer and validate a safe known-good rollback. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

B:

This tempting but unsafe shortcut is not a sufficient remediation for Failed ConnectX HCA or port fault isolation: Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence. A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

C:

That option concerns Intel host CPU PCIe placement and BIOS efficiency. A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU.

Question 1114

At a multi-node AI infrastructure handoff, the team records: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault. Which underlying infrastructure principle BEST explains the issue? Select ONE best answer.

  1. Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence.
  2. A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU.
  3. NVIDIA Xid messages signal GPU-related errors with varying causes and severity; the exact code and context matter.
  4. GPU clock and performance stability depend on sufficient chassis cooling and supported thermal operating envelope.

Explanation

Correct Answer

B:

This is the governing infrastructure principle: A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for Failed ConnectX HCA or port fault isolation: Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence. A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault. A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU.

C:

This alternative is relevant to NVIDIA GPU Xid event root cause isolation but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: NVIDIA Xid messages signal GPU-related errors with varying causes and severity; the exact code and context matter. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

D:

That option concerns Fan faults and thermal throttling investigation. The proportionate action is: Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component.

Question 1115

During an NVIDIA AI factory acceptance test, engineers discover: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault. Which diagnostic evidence should be obtained FIRST to substantiate the likely fault or configuration gap? Select ONE best answer.

  1. Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence.
  2. Review incident timeline, change diffs, hardware alarms, cluster-manager and workload logs, plus rollback comparison.
  3. Inspect lscpu NUMA topology, numactl placement, memory bandwidth, CPU affinity and GPU data-feed stalls.
  4. Compare bidirectional port counters, cable swaps, PCIe link state, adapter logs and test transfers.

Explanation

Correct Answer

D:

This is the direct operational evidence to investigate: Compare bidirectional port counters, cable swaps, PCIe link state, adapter logs and test transfers. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for Failed ConnectX HCA or port fault isolation: Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault. The governing mechanism is: A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU. The proportionate action is: Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component.

B:

This alternative is relevant to Operational fault isolation and safe rollback but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: Review incident timeline, change diffs, hardware alarms, cluster-manager and workload logs, plus rollback comparison. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

C:

This alternative is relevant to AMD host CPU NUMA and memory tuning but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: Inspect lscpu NUMA topology, numactl placement, memory bandwidth, CPU affinity and GPU data-feed stalls. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

Question 1116

A production GPU cluster review identifies this condition: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault. After corrective work, which observable result BEST confirms that the requirement is met? Select ONE best answer.

  1. The repaired node sustains approved clocks and performance with normal fan state under representative load.
  2. The repaired path maintains intended negotiated performance with acceptable errors under repeated RDMA stress.
  3. The device either passes approved health criteria under load or is replaced with traceable fault closure.
  4. Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence.

Explanation

Correct Answer

B:

This is an observable verified result, not merely a claimed configuration change: The repaired path maintains intended negotiated performance with acceptable errors under repeated RDMA stress. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

Incorrect Answers

A:

This alternative is relevant to Fan faults and thermal throttling investigation but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: The repaired node sustains approved clocks and performance with normal fan state under representative load. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

C:

This alternative is relevant to GPU ECC and memory reliability diagnosis but does not directly resolve the current Failed ConnectX HCA or port fault isolation issue: The device either passes approved health criteria under load or is replaced with traceable fault closure. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault.

D:

This tempting but unsafe shortcut is not a sufficient remediation for Failed ConnectX HCA or port fault isolation: Replace healthy GPU cards to fix every InfiniBand timeout without inspecting NIC ports and fiber error evidence. Observed scenario: A team replaces a whole GPU node before testing the transceiver and cable responsible for its RDMA fault. The governing mechanism is: A high-speed fabric fault may originate in cable, transceiver, adapter, driver or host slot rather than the GPU. The proportionate action is: Localize HCA, port, optic, cabling and PCIe causes before replacing the approved failed component. A direct check is: Compare bidirectional port counters, cable swaps, PCIe link state, adapter logs and test transfers. The acceptance condition is: The repaired path maintains intended negotiated performance with acceptable errors under repeated RDMA stress.

Question 1117

In a professional AI infrastructure deployment, a technician observes: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs. Which operational response is MOST appropriate to address the root issue? Select ONE best answer.

  1. Assume any physically fitting GPU board can be hot-swapped into every running NVIDIA HGX system without validation.
  2. Drain workloads and follow the vendor's approved component replacement, firmware and postrepair diagnostic runbook.
  3. Isolate the failed power path, verify redundancy budget and follow supported PSU replacement with monitoring.
  4. Profile representative file access, mount options, caches and pipeline parallelism before tuning or scaling storage.

Explanation

Correct Answer

B:

This is an appropriate correction for GPU and NVSwitch field replaceable component workflow: Drain workloads and follow the vendor's approved component replacement, firmware and postrepair diagnostic runbook. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for GPU and NVSwitch field replaceable component workflow: Assume any physically fitting GPU board can be hot-swapped into every running NVIDIA HGX system without validation.

C:

That option concerns Redundant PSU and power path failure replacement. GPU and switch field replacement changes hardware identity and fabric behavior and requires safety and validation steps.

D:

That option concerns Storage mounts latency and data pipeline tuning. GPU and switch field replacement changes hardware identity and fabric behavior and requires safety and validation steps.

Question 1118

The cluster operations team needs to evaluate this situation: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs. Which underlying infrastructure principle BEST explains the issue? Select ONE best answer.

  1. Server CPU socket topology, PCIe lane allocation and memory access shape the efficiency of host-to-GPU transfers.
  2. GPU and switch field replacement changes hardware identity and fabric behavior and requires safety and validation steps.
  3. GPU clock and performance stability depend on sufficient chassis cooling and supported thermal operating envelope.
  4. Assume any physically fitting GPU board can be hot-swapped into every running NVIDIA HGX system without validation.

Explanation

Correct Answer

B:

This is the governing infrastructure principle: GPU and switch field replacement changes hardware identity and fabric behavior and requires safety and validation steps. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs.

Incorrect Answers

A:

That option concerns Intel host CPU PCIe placement and BIOS efficiency. A direct check is: Inspect asset identifiers, FRU service records, firmware compatibility, PCIe enumeration and NVLink or NCCL results.

C:

That option concerns Fan faults and thermal throttling investigation. A direct check is: Inspect asset identifiers, FRU service records, firmware compatibility, PCIe enumeration and NVLink or NCCL results.

D:

This tempting but unsafe shortcut is not a sufficient remediation for GPU and NVSwitch field replaceable component workflow: Assume any physically fitting GPU board can be hot-swapped into every running NVIDIA HGX system without validation. GPU and switch field replacement changes hardware identity and fabric behavior and requires safety and validation steps.

Question 1119

At a multi-node AI infrastructure handoff, the team records: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs. Which diagnostic evidence should be obtained FIRST to substantiate the likely fault or configuration gap? Select ONE best answer.

  1. Inspect asset identifiers, FRU service records, firmware compatibility, PCIe enumeration and NVLink or NCCL results.
  2. Inspect fan tachometer and BMC sensors, GPU temperature, throttle reasons, inlet mapping and workload timing.
  3. Assume any physically fitting GPU board can be hot-swapped into every running NVIDIA HGX system without validation.
  4. Inspect lscpu NUMA topology, numactl placement, memory bandwidth, CPU affinity and GPU data-feed stalls.

Explanation

Correct Answer

A:

This is the direct operational evidence to investigate: Inspect asset identifiers, FRU service records, firmware compatibility, PCIe enumeration and NVLink or NCCL results. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs.

Incorrect Answers

B:

This alternative is relevant to Fan faults and thermal throttling investigation but does not directly resolve the current GPU and NVSwitch field replaceable component workflow issue: Inspect fan tachometer and BMC sensors, GPU temperature, throttle reasons, inlet mapping and workload timing. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs.

C:

This tempting but unsafe shortcut is not a sufficient remediation for GPU and NVSwitch field replaceable component workflow: Assume any physically fitting GPU board can be hot-swapped into every running NVIDIA HGX system without validation. An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs.

D:

This alternative is relevant to AMD host CPU NUMA and memory tuning but does not directly resolve the current GPU and NVSwitch field replaceable component workflow issue: Inspect lscpu NUMA topology, numactl placement, memory bandwidth, CPU affinity and GPU data-feed stalls. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs.

Question 1120

During an NVIDIA AI factory acceptance test, engineers discover: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs. Which TWO complementary elements are needed: the correct technical action and the strongest independent diagnostic evidence? Select TWO answers.

  1. Review kernel and NVIDIA Xid logs, driver version, concurrent workload, ECC state and vendor diagnostic guidance.
  2. Assume any physically fitting GPU board can be hot-swapped into every running NVIDIA HGX system without validation.
  3. Validate Intel CPU, PCIe and NUMA placement with supported BIOS and stable sustained-power settings.
  4. Inspect asset identifiers, FRU service records, firmware compatibility, PCIe enumeration and NVLink or NCCL results.
  5. Drain workloads and follow the vendor's approved component replacement, firmware and postrepair diagnostic runbook.

Explanation

Correct Answers

D:

This is the direct operational evidence to investigate: Inspect asset identifiers, FRU service records, firmware compatibility, PCIe enumeration and NVLink or NCCL results. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs. The governing mechanism is: GPU and switch field replacement changes hardware identity and fabric behavior and requires safety and validation steps.

E:

This is an appropriate correction for GPU and NVSwitch field replaceable component workflow: Drain workloads and follow the vendor's approved component replacement, firmware and postrepair diagnostic runbook. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs. The governing mechanism is: GPU and switch field replacement changes hardware identity and fabric behavior and requires safety and validation steps.

Incorrect Answers

A:

This alternative is relevant to NVIDIA GPU Xid event root cause isolation but does not directly resolve the current GPU and NVSwitch field replaceable component workflow issue: Review kernel and NVIDIA Xid logs, driver version, concurrent workload, ECC state and vendor diagnostic guidance. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs.

B:

This tempting but unsafe shortcut is not a sufficient remediation for GPU and NVSwitch field replaceable component workflow: Assume any physically fitting GPU board can be hot-swapped into every running NVIDIA HGX system without validation. An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs. GPU and switch field replacement changes hardware identity and fabric behavior and requires safety and validation steps.

C:

This alternative is relevant to Intel host CPU PCIe placement and BIOS efficiency but does not directly resolve the current GPU and NVSwitch field replaceable component workflow issue: Validate Intel CPU, PCIe and NUMA placement with supported BIOS and stable sustained-power settings. Observed scenario: An HGX tray reports one defective accelerator and a technician plans service without draining active training jobs.

NCP-AI Infrastructure (NCP-AII) Practice Tests