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 52

Question 510

During an NVIDIA AI factory acceptance test, engineers discover: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device. Which TWO complementary elements are needed: the correct technical action and the strongest independent diagnostic evidence? Select TWO answers.

  1. Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured.
  2. Define supported Slurm controller, compute nodes, partitions and authorized scheduling policies together.
  3. Inspect slurmd discovery, gres configuration, CUDA_VISIBLE_DEVICES and simultaneous job placement.
  4. Align supported interface or bond configuration, RDMA stack and fabric QoS with the cluster topology.
  5. Inspect interfaces, RDMA devices, routing, bond membership, link counters and representative transfer tests.

Explanation

Correct Answers

D:

This is an appropriate correction for Compute fabric interface bonding and RDMA prerequisites: Align supported interface or bond configuration, RDMA stack and fabric QoS with the cluster topology. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device. The governing mechanism is: IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment.

E:

This is the direct operational evidence to investigate: Inspect interfaces, RDMA devices, routing, bond membership, link counters and representative transfer tests. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for Compute fabric interface bonding and RDMA prerequisites: Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured. An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device. IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment.

B:

This alternative is relevant to Slurm controller partition setup but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Define supported Slurm controller, compute nodes, partitions and authorized scheduling policies together. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device.

C:

This alternative is relevant to Slurm GPU GRES enumeration and device mapping but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Inspect slurmd discovery, gres configuration, CUDA_VISIBLE_DEVICES and simultaneous job placement. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device.

Question 511

A production GPU cluster review identifies this condition: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device. Which diagnostic evidence should be obtained FIRST to substantiate the likely fault or configuration gap? Select ONE best answer.

  1. Review authorized key and account policy, SSH server settings, firewall reachability and authentication logs.
  2. Inspect category membership, inherited images, node overrides, resulting configuration and change impact.
  3. Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured.
  4. Inspect interfaces, RDMA devices, routing, bond membership, link counters and representative transfer tests.

Explanation

Correct Answer

D:

This is the direct operational evidence to investigate: Inspect interfaces, RDMA devices, routing, bond membership, link counters and representative transfer tests. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device. The governing mechanism is: IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment.

Incorrect Answers

A:

This alternative is relevant to Host SSH certificates keys and OS access segregation but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Review authorized key and account policy, SSH server settings, firewall reachability and authentication logs. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device.

B:

This alternative is relevant to BCM node categories and role inheritance but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Inspect category membership, inherited images, node overrides, resulting configuration and change impact. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device.

C:

This tempting but unsafe shortcut is not a sufficient remediation for Compute fabric interface bonding and RDMA prerequisites: Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device. The governing mechanism is: IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment. The proportionate action is: Align supported interface or bond configuration, RDMA stack and fabric QoS with the cluster topology.

Question 512

In a professional AI infrastructure deployment, a technician observes: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device. After corrective work, which observable result BEST confirms that the requirement is met? Select ONE best answer.

  1. Each workload reaches the intended high-speed path and RDMA transfer passes with correct device selection.
  2. Required RDMA and DOCA capabilities operate on canary and production nodes without incompatible modules.
  3. The intended worker reliably reboots into the approved OS and image with the right management identity.
  4. Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured.

Explanation

Correct Answer

A:

This is an observable verified result, not merely a claimed configuration change: Each workload reaches the intended high-speed path and RDMA transfer passes with correct device selection. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device.

Incorrect Answers

B:

This alternative is relevant to DOCA host networking and RDMA driver stack but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Required RDMA and DOCA capabilities operate on canary and production nodes without incompatible modules. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device.

C:

This alternative is relevant to BCM image provisioning and PXE worker boot but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: The intended worker reliably reboots into the approved OS and image with the right management identity. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device.

D:

This tempting but unsafe shortcut is not a sufficient remediation for Compute fabric interface bonding and RDMA prerequisites: Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured. Observed scenario: An InfiniBand worker has the correct IP but its RDMA userspace libraries cannot open a device. The governing mechanism is: IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment. The proportionate action is: Align supported interface or bond configuration, RDMA stack and fabric QoS with the cluster topology. A direct check is: Inspect interfaces, RDMA devices, routing, bond membership, link counters and representative transfer tests. The acceptance condition is: Each workload reaches the intended high-speed path and RDMA transfer passes with correct device selection.

Question 513

The cluster operations team needs to evaluate this situation: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade. Which operational response is MOST appropriate to address the root issue? Select ONE best answer.

  1. Align supported interface or bond configuration, RDMA stack and fabric QoS with the cluster topology.
  2. Enforce individual authenticated host access, network restrictions and tested key rotation with accountable emergency recovery.
  3. Use NVIDIA-supported OFED or DOCA-host and driver packages and verify host-DPU compatibility before promotion.
  4. Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured.

Explanation

Correct Answer

A:

This is an appropriate correction for Compute fabric interface bonding and RDMA prerequisites: Align supported interface or bond configuration, RDMA stack and fabric QoS with the cluster topology. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

Incorrect Answers

B:

This alternative is relevant to Host SSH certificates keys and OS access segregation but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Enforce individual authenticated host access, network restrictions and tested key rotation with accountable emergency recovery. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

C:

This alternative is relevant to DOCA host networking and RDMA driver stack but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Use NVIDIA-supported OFED or DOCA-host and driver packages and verify host-DPU compatibility before promotion. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

D:

This tempting but unsafe shortcut is not a sufficient remediation for Compute fabric interface bonding and RDMA prerequisites: Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured. An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

Question 514

At a multi-node AI infrastructure handoff, the team records: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade. Which underlying infrastructure principle BEST explains the issue? Select ONE best answer.

  1. Accelerator and fabric software are kernel-dependent; a generic successful Linux boot does not establish full GPU readiness.
  2. Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured.
  3. Cluster managers coordinate images, nodes, services and configuration; worker OS boot cannot replace control-plane readiness.
  4. IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment.

Explanation

Correct Answer

D:

This is the governing infrastructure principle: IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

Incorrect Answers

A:

That option concerns Linux operating system and kernel module support. IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment.

B:

This tempting but unsafe shortcut is not a sufficient remediation for Compute fabric interface bonding and RDMA prerequisites: Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured. An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade. IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment.

C:

This alternative is relevant to Base Command Manager head-node installation but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Cluster managers coordinate images, nodes, services and configuration; worker OS boot cannot replace control-plane readiness. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

Question 515

During an NVIDIA AI factory acceptance test, engineers discover: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade. Which diagnostic evidence should be obtained FIRST to substantiate the likely fault or configuration gap? Select ONE best answer.

  1. Review slurm.conf, controller health, partition listings, node state and a submitted sample job.
  2. Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured.
  3. Inspect interfaces, RDMA devices, routing, bond membership, link counters and representative transfer tests.
  4. Inspect DHCP leases, PXE requests, image revisions, BCM node mappings and host boot traces.

Explanation

Correct Answer

C:

This is the direct operational evidence to investigate: Inspect interfaces, RDMA devices, routing, bond membership, link counters and representative transfer tests. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

Incorrect Answers

A:

This alternative is relevant to Slurm controller partition setup but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Review slurm.conf, controller health, partition listings, node state and a submitted sample job. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

B:

This tempting but unsafe shortcut is not a sufficient remediation for Compute fabric interface bonding and RDMA prerequisites: Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade. The governing mechanism is: IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment. The proportionate action is: Align supported interface or bond configuration, RDMA stack and fabric QoS with the cluster topology.

D:

This alternative is relevant to BCM image provisioning and PXE worker boot but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: Inspect DHCP leases, PXE requests, image revisions, BCM node mappings and host boot traces. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

Question 516

A production GPU cluster review identifies this condition: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade. After corrective work, which observable result BEST confirms that the requirement is met? Select ONE best answer.

  1. Each workload reaches the intended high-speed path and RDMA transfer passes with correct device selection.
  2. A planned primary outage transfers management to standby without losing cluster configuration integrity.
  3. The management node reports correct cluster state and consistently deploys an approved image to intended nodes.
  4. Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured.

Explanation

Correct Answer

A:

This is an observable verified result, not merely a claimed configuration change: Each workload reaches the intended high-speed path and RDMA transfer passes with correct device selection. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

Incorrect Answers

B:

That option concerns BCM high availability and management failover. IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment.

C:

This alternative is relevant to Base Command Manager head-node installation but does not directly resolve the current Compute fabric interface bonding and RDMA prerequisites issue: The management node reports correct cluster state and consistently deploys an approved image to intended nodes. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade.

D:

This tempting but unsafe shortcut is not a sufficient remediation for Compute fabric interface bonding and RDMA prerequisites: Assume successful ICMP ping proves that InfiniBand verbs and RoCE queues are fully configured. Observed scenario: An Ethernet RoCE node misclassifies priority traffic after a VLAN change and storage transfers degrade. The governing mechanism is: IP connectivity does not prove correct RDMA capability, link aggregation or loss and congestion treatment. The proportionate action is: Align supported interface or bond configuration, RDMA stack and fabric QoS with the cluster topology. A direct check is: Inspect interfaces, RDMA devices, routing, bond membership, link counters and representative transfer tests. The acceptance condition is: Each workload reaches the intended high-speed path and RDMA transfer passes with correct device selection.

Question 517

In a professional AI infrastructure deployment, a technician observes: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs. Which operational response is MOST appropriate to address the root issue? Select ONE best answer.

  1. Define explicit BCM node categories and managed overrides for shared hardware and service configuration.
  2. Install the supported BCM head-node services and preserve configuration data before provisioning and accepting workers.
  3. Define supported Slurm controller, compute nodes, partitions and authorized scheduling policies together.
  4. Declare a Slurm cluster ready because slurmd starts, even if there is no usable GPU partition or active controller.

Explanation

Correct Answer

C:

This is an appropriate correction for Slurm controller partition setup: Define supported Slurm controller, compute nodes, partitions and authorized scheduling policies together. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

Incorrect Answers

A:

This alternative is relevant to BCM node categories and role inheritance but does not directly resolve the current Slurm controller partition setup issue: Define explicit BCM node categories and managed overrides for shared hardware and service configuration. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

B:

This alternative is relevant to Base Command Manager head-node installation but does not directly resolve the current Slurm controller partition setup issue: Install the supported BCM head-node services and preserve configuration data before provisioning and accepting workers. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

D:

This tempting but unsafe shortcut is not a sufficient remediation for Slurm controller partition setup: Declare a Slurm cluster ready because slurmd starts, even if there is no usable GPU partition or active controller.

Question 518

The cluster operations team needs to evaluate this situation: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs. Which underlying infrastructure principle BEST explains the issue? Select ONE best answer.

  1. Schedulers allocate declared GPU resources; a mismatch between configured GRES and detected devices causes wrong placement.
  2. A working Slurm daemon is not enough; partitions determine which nodes and resources can serve jobs.
  3. Declare a Slurm cluster ready because slurmd starts, even if there is no usable GPU partition or active controller.
  4. Adding a second management host does not prove cluster-manager continuity unless state and advertised services fail over.

Explanation

Correct Answer

B:

This is the governing infrastructure principle: A working Slurm daemon is not enough; partitions determine which nodes and resources can serve jobs. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

Incorrect Answers

A:

This alternative is relevant to Slurm GPU GRES enumeration and device mapping but does not directly resolve the current Slurm controller partition setup issue: Schedulers allocate declared GPU resources; a mismatch between configured GRES and detected devices causes wrong placement. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

C:

This tempting but unsafe shortcut is not a sufficient remediation for Slurm controller partition setup: Declare a Slurm cluster ready because slurmd starts, even if there is no usable GPU partition or active controller. A working Slurm daemon is not enough; partitions determine which nodes and resources can serve jobs.

D:

This alternative is relevant to BCM high availability and management failover but does not directly resolve the current Slurm controller partition setup issue: Adding a second management host does not prove cluster-manager continuity unless state and advertised services fail over. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

Question 519

At a multi-node AI infrastructure handoff, the team records: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs. Which diagnostic evidence should be obtained FIRST to substantiate the likely fault or configuration gap? Select ONE best answer.

  1. Inspect BCM service health, database backups, node discovery, management interfaces and provisioning records.
  2. Declare a Slurm cluster ready because slurmd starts, even if there is no usable GPU partition or active controller.
  3. Review slurm.conf, controller health, partition listings, node state and a submitted sample job.
  4. Inspect slurmd discovery, gres configuration, CUDA_VISIBLE_DEVICES and simultaneous job placement.

Explanation

Correct Answer

C:

This is the direct operational evidence to investigate: Review slurm.conf, controller health, partition listings, node state and a submitted sample job. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

Incorrect Answers

A:

This alternative is relevant to Base Command Manager head-node installation but does not directly resolve the current Slurm controller partition setup issue: Inspect BCM service health, database backups, node discovery, management interfaces and provisioning records. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

B:

This tempting but unsafe shortcut is not a sufficient remediation for Slurm controller partition setup: Declare a Slurm cluster ready because slurmd starts, even if there is no usable GPU partition or active controller. A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

D:

This alternative is relevant to Slurm GPU GRES enumeration and device mapping but does not directly resolve the current Slurm controller partition setup issue: Inspect slurmd discovery, gres configuration, CUDA_VISIBLE_DEVICES and simultaneous job placement. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

Question 520

During an NVIDIA AI factory acceptance test, engineers discover: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs. Which TWO complementary elements are needed: the correct technical action and the strongest independent diagnostic evidence? Select TWO answers.

  1. Define supported Slurm controller, compute nodes, partitions and authorized scheduling policies together.
  2. Declare a Slurm cluster ready because slurmd starts, even if there is no usable GPU partition or active controller.
  3. Review slurm.conf, controller health, partition listings, node state and a submitted sample job.
  4. Inspect partition time limits, QOS priorities, pending reasons, preemption settings and maintenance reservations.
  5. Install the supported BCM head-node services and preserve configuration data before provisioning and accepting workers.

Explanation

Correct Answers

A:

This is an appropriate correction for Slurm controller partition setup: Define supported Slurm controller, compute nodes, partitions and authorized scheduling policies together. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs. The governing mechanism is: A working Slurm daemon is not enough; partitions determine which nodes and resources can serve jobs.

C:

This is the direct operational evidence to investigate: Review slurm.conf, controller health, partition listings, node state and a submitted sample job. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs. The governing mechanism is: A working Slurm daemon is not enough; partitions determine which nodes and resources can serve jobs.

Incorrect Answers

B:

This tempting but unsafe shortcut is not a sufficient remediation for Slurm controller partition setup: Declare a Slurm cluster ready because slurmd starts, even if there is no usable GPU partition or active controller. A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs. A working Slurm daemon is not enough; partitions determine which nodes and resources can serve jobs.

D:

This alternative is relevant to Slurm scheduling time limits fairness and preemption but does not directly resolve the current Slurm controller partition setup issue: Inspect partition time limits, QOS priorities, pending reasons, preemption settings and maintenance reservations. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs.

E:

This alternative is relevant to Base Command Manager head-node installation but does not directly resolve the current Slurm controller partition setup issue: Install the supported BCM head-node services and preserve configuration data before provisioning and accepting workers. Observed scenario: A cluster runs slurmd on workers but no configured controller partition can accept GPU training jobs. The governing mechanism is: A working Slurm daemon is not enough; partitions determine which nodes and resources can serve jobs.

NCP-AI Infrastructure (NCP-AII) Practice Tests