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 22

Question 210

An infrastructure administrator receives the following incident: A team reconfigures MIG geometry while production GPU jobs are active, causing disruptive device recreation and lost workloads. Which measured result most directly proves that this specific problem is resolved?

  1. A device inventory and representative job test show the intended MIG instances, resource quotas, and successful workload placement.
  2. After a controlled partition reconfiguration, the intended job maps to the approved GPU and MIG UUID with expected memory capacity.
  3. A controlled test shows the traffic follows the documented DPU enforcement path and host management and datapath visibility agree on the mode.
  4. A staged reconfiguration returns the target MIG profile set, and no protected job experiences an unplanned partition reset.

Explanation

Correct Answer

D:

This is the measurable acceptance result for the described case: A staged reconfiguration returns the target MIG profile set, and no protected job experiences an unplanned partition reset. It validates the corrective operation (Drain affected workloads and use the documented enable/disable and instance creation sequence before returning capacity to service.) rather than a related but different system check.

Incorrect Answers

A:

This observation would help verify a different incident: Researchers enable Multi-Instance GPU on a supported accelerator but assume two slices share the same capacity and hardware topology as one full GPU. It does not directly establish the required result for the current case: A staged reconfiguration returns the target MIG profile set, and no protected job experiences an unplanned partition reset.

B:

This observation would help verify a different incident: A scheduler assigns a training container to a MIG instance by a temporary index, and reconfiguration leaves the container using the wrong partition. It does not directly establish the required result for the current case: A staged reconfiguration returns the target MIG profile set, and no protected job experiences an unplanned partition reset.

C:

This observation would help verify a different incident: A GPU rack uses BlueField adapters with an unexpected embedded CPU management mode, and host traffic bypasses the data-plane policies administrators expected on the DPU. It does not directly establish the required result for the current case: A staged reconfiguration returns the target MIG profile set, and no protected job experiences an unplanned partition reset.

Question 211

An infrastructure administrator receives the following incident: A team reconfigures MIG geometry while production GPU jobs are active, causing disruptive device recreation and lost workloads. Which root cause or infrastructure principle best explains the observed failure?

  1. MIG creates bounded GPU compute and memory instances rather than several unrestricted copies of the full physical accelerator.
  2. Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed.
  3. MIG topology changes can invalidate ephemeral index-based device placement and monitoring assumptions.
  4. BlueField operating modes change which processor owns network and infrastructure functions, so a physical adapter alone does not guarantee the intended services.

Explanation

Correct Answer

B:

This technical mechanism explains the reported condition: Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed. The observed situation is: A team reconfigures MIG geometry while production GPU jobs are active, causing disruptive device recreation and lost workloads. A requirement-aligned correction is: Drain affected workloads and use the documented enable/disable and instance creation sequence before returning capacity to service.

Incorrect Answers

A:

This underlying cause is associated with another GPU platform incident: Researchers enable Multi-Instance GPU on a supported accelerator but assume two slices share the same capacity and hardware topology as one full GPU. The observed case instead concerns: Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed.

C:

This underlying cause is associated with another GPU platform incident: A scheduler assigns a training container to a MIG instance by a temporary index, and reconfiguration leaves the container using the wrong partition. The observed case instead concerns: Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed.

D:

This underlying cause is associated with another GPU platform incident: A GPU rack uses BlueField adapters with an unexpected embedded CPU management mode, and host traffic bypasses the data-plane policies administrators expected on the DPU. The observed case instead concerns: Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed.

Question 212

An infrastructure administrator receives the following incident: A team reconfigures MIG geometry while production GPU jobs are active, causing disruptive device recreation and lost workloads. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Validate supported MIG geometry and workload sizing before offering shared accelerator partitions to tenants.
  2. Require device rediscovery, allocation reconciliation, and job-drain checks whenever MIG partitions are reconfigured.
  3. Enforce maintenance windows, scheduler cordoning, and before/after MIG inventory checks for every layout change.
  4. Gate DPU firmware rollout on signed images, mode compatibility, canary validation, and a tested recovery path.

Explanation

Correct Answer

C:

This preventive practice addresses the underlying failure (Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed.) and supports continued assurance through: A staged reconfiguration returns the target MIG profile set, and no protected job experiences an unplanned partition reset.

Incorrect Answers

A:

This prevention control targets another operational risk: Researchers enable Multi-Instance GPU on a supported accelerator but assume two slices share the same capacity and hardware topology as one full GPU. The currently reported mechanism remains: Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed.

B:

This prevention control targets another operational risk: A scheduler assigns a training container to a MIG instance by a temporary index, and reconfiguration leaves the container using the wrong partition. The currently reported mechanism remains: Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed.

D:

This prevention control targets another operational risk: A security team changes BlueField firmware on all AI servers but does not retain a recovery image, and one canary device fails to complete boot. The currently reported mechanism remains: Changing physical GPU partitioning can invalidate live device allocations even when the host driver remains installed.

Question 213

An NVIDIA accelerator platform readiness review identifies this issue: Base Command Manager is installed on a management node but the cluster database is stored only on a single local disk, and leadership claims the control plane is highly available. Which corrective deployment or operational action best addresses the stated problem?

  1. Configure supported HA quorum or fencing behavior and verify that only one authoritative control-plane leader can perform changes.
  2. Rehearse the version change with representative compute nodes and scheduling integrations and execute a staged upgrade with rollback.
  3. Design and verify BCM high availability with supported redundant management services and resilient state storage.
  4. Initialize the supported NGC client configuration with an authorized organization-scoped credential and verify registry access.

Explanation

Correct Answer

C:

For the stated NVIDIA AI infrastructure condition (Base Command Manager is installed on a management node but the cluster database is stored only on a single local disk, and leadership claims the control plane is highly available.), this action addresses the concrete failure: Design and verify BCM high availability with supported redundant management services and resilient state storage. The underlying mechanism is: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk. Verification would require: A controlled loss of the primary management node preserves or restores administrative provisioning through the supported standby within the accepted recovery target.

Incorrect Answers

A:

This option could be appropriate for a different NVIDIA infrastructure problem (3.1: Install Base Command Manager and configure and verify high availability): An AI cluster has a functioning BCM primary and standby, but a split-brain test reveals that both managers can modify node provisioning state after losing their heartbeat. The present issue is more specifically caused by: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

B:

This option could be appropriate for a different NVIDIA infrastructure problem (3.1: Install Base Command Manager and configure and verify high availability): An operations team upgrades BCM directly in production without testing whether existing Slurm and provisioning integrations are compatible with the new management API. The present issue is more specifically caused by: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

D:

This option could be appropriate for a different NVIDIA infrastructure problem (3.7: Install NGC CLI on hosts): An NGC CLI client is installed on a shared administrative workstation but lacks authenticated registry configuration, so container downloads fail with permissions errors. The present issue is more specifically caused by: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

Question 214

An NVIDIA accelerator platform readiness review identifies this issue: Base Command Manager is installed on a management node but the cluster database is stored only on a single local disk, and leadership claims the control plane is highly available. Which measured result most directly proves that this specific problem is resolved?

  1. A controlled loss of the primary management node preserves or restores administrative provisioning through the supported standby within the accepted recovery target.
  2. A controlled network boot obtains the correct lease and approved image and enrolls the node under the expected BCM identity.
  3. Reimaged target nodes join the proper GPU resource category and expose the expected driver and network interface baseline.
  4. The signed module loads under the enforced boot policy and a CUDA smoke test completes without reducing platform integrity controls.

Explanation

Correct Answer

A:

This is the measurable acceptance result for the described case: A controlled loss of the primary management node preserves or restores administrative provisioning through the supported standby within the accepted recovery target. It validates the corrective operation (Design and verify BCM high availability with supported redundant management services and resilient state storage.) rather than a related but different system check.

Incorrect Answers

B:

This observation would help verify a different incident: A newly added AI rack cannot be provisioned because BCM is reachable on the data network but its PXE/DHCP provisioning path is isolated by a recently changed VLAN. It does not directly establish the required result for the current case: A controlled loss of the primary management node preserves or restores administrative provisioning through the supported standby within the accepted recovery target.

C:

This observation would help verify a different incident: A BCM provisioning environment discovers every compute node, yet new GPU hosts join the wrong category and receive CPU-only operating system images. It does not directly establish the required result for the current case: A controlled loss of the primary management node preserves or restores administrative provisioning through the supported standby within the accepted recovery target.

D:

This observation would help verify a different incident: A security patch changes the host kernel and Secure Boot settings, and the NVIDIA GPU module stops loading despite the device remaining visible on PCIe. It does not directly establish the required result for the current case: A controlled loss of the primary management node preserves or restores administrative provisioning through the supported standby within the accepted recovery target.

Question 215

An NVIDIA accelerator platform readiness review identifies this issue: Base Command Manager is installed on a management node but the cluster database is stored only on a single local disk, and leadership claims the control plane is highly available. Which root cause or infrastructure principle best explains the observed failure?

  1. Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.
  2. Cluster management upgrades can affect downstream automation interfaces even when the central manager starts normally.
  3. Hardware discovery alone cannot ensure workload readiness when the control plane applies an incorrect provisioning profile.
  4. Installing a registry client does not configure identity, organizational entitlement, or secure token handling.

Explanation

Correct Answer

A:

This technical mechanism explains the reported condition: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk. The observed situation is: Base Command Manager is installed on a management node but the cluster database is stored only on a single local disk, and leadership claims the control plane is highly available. A requirement-aligned correction is: Design and verify BCM high availability with supported redundant management services and resilient state storage.

Incorrect Answers

B:

This underlying cause is associated with another GPU platform incident: An operations team upgrades BCM directly in production without testing whether existing Slurm and provisioning integrations are compatible with the new management API. The observed case instead concerns: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

C:

This underlying cause is associated with another GPU platform incident: A BCM provisioning environment discovers every compute node, yet new GPU hosts join the wrong category and receive CPU-only operating system images. The observed case instead concerns: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

D:

This underlying cause is associated with another GPU platform incident: An NGC CLI client is installed on a shared administrative workstation but lacks authenticated registry configuration, so container downloads fail with permissions errors. The observed case instead concerns: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

Question 216

An NVIDIA accelerator platform readiness review identifies this issue: Base Command Manager is installed on a management node but the cluster database is stored only on a single local disk, and leadership claims the control plane is highly available. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Review and rehearse manager failover, data backup, and state consistency as part of BCM release acceptance.
  2. Include leader isolation, fencing, and database recovery in each BCM high-availability test plan.
  3. Audit category inheritance, image mapping, and node labels whenever accelerator hardware types are added.
  4. Record container CUDA requirements and enforce driver-capability checks during rollout.

Explanation

Correct Answer

A:

This preventive practice addresses the underlying failure (Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.) and supports continued assurance through: A controlled loss of the primary management node preserves or restores administrative provisioning through the supported standby within the accepted recovery target.

Incorrect Answers

B:

This prevention control targets another operational risk: An AI cluster has a functioning BCM primary and standby, but a split-brain test reveals that both managers can modify node provisioning state after losing their heartbeat. The currently reported mechanism remains: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

C:

This prevention control targets another operational risk: A BCM provisioning environment discovers every compute node, yet new GPU hosts join the wrong category and receive CPU-only operating system images. The currently reported mechanism remains: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

D:

This prevention control targets another operational risk: A container sees its host GPU but fails when run on a different node whose driver is older than the image's required CUDA feature level. The currently reported mechanism remains: Two management host names do not establish HA if control-plane state remains on one unrecoverable local disk.

Question 217

During a GPU datacenter maintenance window, an engineer discovers: An AI cluster has a functioning BCM primary and standby, but a split-brain test reveals that both managers can modify node provisioning state after losing their heartbeat. Which corrective deployment or operational action best addresses the stated problem?

  1. Rehearse the version change with representative compute nodes and scheduling integrations and execute a staged upgrade with rollback.
  2. Configure supported HA quorum or fencing behavior and verify that only one authoritative control-plane leader can perform changes.
  3. Correct image and node-category assignments in the supported BCM management configuration before provisioning the affected hosts.
  4. Deploy versioned golden images or controlled image automation and reconcile host drift before promoting updates.

Explanation

Correct Answer

B:

For the stated NVIDIA AI infrastructure condition (An AI cluster has a functioning BCM primary and standby, but a split-brain test reveals that both managers can modify node provisioning state after losing their heartbeat.), this action addresses the concrete failure: Configure supported HA quorum or fencing behavior and verify that only one authoritative control-plane leader can perform changes. The underlying mechanism is: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions. Verification would require: A forced partition leaves a single active management authority and protects configuration state against conflicting writes.

Incorrect Answers

A:

This option could be appropriate for a different NVIDIA infrastructure problem (3.1: Install Base Command Manager and configure and verify high availability): An operations team upgrades BCM directly in production without testing whether existing Slurm and provisioning integrations are compatible with the new management API. The present issue is more specifically caused by: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

C:

This option could be appropriate for a different NVIDIA infrastructure problem (3.1: Install Base Command Manager and configure and verify high availability): A BCM provisioning environment discovers every compute node, yet new GPU hosts join the wrong category and receive CPU-only operating system images. The present issue is more specifically caused by: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

D:

This option could be appropriate for a different NVIDIA infrastructure problem (3.2: Install operating systems): A fleet of GPU hosts uses manually customized operating system packages, making a failed driver upgrade impossible to reproduce or roll back consistently. The present issue is more specifically caused by: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

Question 218

During a GPU datacenter maintenance window, an engineer discovers: An AI cluster has a functioning BCM primary and standby, but a split-brain test reveals that both managers can modify node provisioning state after losing their heartbeat. Which measured result most directly proves that this specific problem is resolved?

  1. A controlled network boot obtains the correct lease and approved image and enrolls the node under the expected BCM identity.
  2. A forced partition leaves a single active management authority and protects configuration state against conflicting writes.
  3. A controlled loss of the primary management node preserves or restores administrative provisioning through the supported standby within the accepted recovery target.
  4. A denied test account cannot schedule into the restricted partition while an approved account receives its expected GPU allocation.

Explanation

Correct Answer

B:

This is the measurable acceptance result for the described case: A forced partition leaves a single active management authority and protects configuration state against conflicting writes. It validates the corrective operation (Configure supported HA quorum or fencing behavior and verify that only one authoritative control-plane leader can perform changes.) rather than a related but different system check.

Incorrect Answers

A:

This observation would help verify a different incident: A newly added AI rack cannot be provisioned because BCM is reachable on the data network but its PXE/DHCP provisioning path is isolated by a recently changed VLAN. It does not directly establish the required result for the current case: A forced partition leaves a single active management authority and protects configuration state against conflicting writes.

C:

This observation would help verify a different incident: Base Command Manager is installed on a management node but the cluster database is stored only on a single local disk, and leadership claims the control plane is highly available. It does not directly establish the required result for the current case: A forced partition leaves a single active management authority and protects configuration state against conflicting writes.

D:

This observation would help verify a different incident: A new node is added to the wrong Slurm partition and gains access to restricted customer workloads despite being intended for untrusted experimentation. It does not directly establish the required result for the current case: A forced partition leaves a single active management authority and protects configuration state against conflicting writes.

Question 219

During a GPU datacenter maintenance window, an engineer discovers: An AI cluster has a functioning BCM primary and standby, but a split-brain test reveals that both managers can modify node provisioning state after losing their heartbeat. Which root cause or infrastructure principle best explains the observed failure?

  1. Ordinary manager UI reachability does not prove that low-level network boot and imaging services can reach compute hosts.
  2. Cluster management upgrades can affect downstream automation interfaces even when the central manager starts normally.
  3. An installed host driver does not automatically expose accelerators inside isolated containers.
  4. Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

Explanation

Correct Answer

D:

This technical mechanism explains the reported condition: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions. The observed situation is: An AI cluster has a functioning BCM primary and standby, but a split-brain test reveals that both managers can modify node provisioning state after losing their heartbeat. A requirement-aligned correction is: Configure supported HA quorum or fencing behavior and verify that only one authoritative control-plane leader can perform changes.

Incorrect Answers

A:

This underlying cause is associated with another GPU platform incident: A newly added AI rack cannot be provisioned because BCM is reachable on the data network but its PXE/DHCP provisioning path is isolated by a recently changed VLAN. The observed case instead concerns: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

B:

This underlying cause is associated with another GPU platform incident: An operations team upgrades BCM directly in production without testing whether existing Slurm and provisioning integrations are compatible with the new management API. The observed case instead concerns: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

C:

This underlying cause is associated with another GPU platform incident: A Docker AI image launches on a GPU host but reports no CUDA devices because the NVIDIA container runtime integration was never installed. The observed case instead concerns: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

Question 220

During a GPU datacenter maintenance window, an engineer discovers: An AI cluster has a functioning BCM primary and standby, but a split-brain test reveals that both managers can modify node provisioning state after losing their heartbeat. Which repeatable operational safeguard best prevents recurrence of this failure?

  1. Include leader isolation, fencing, and database recovery in each BCM high-availability test plan.
  2. Audit category inheritance, image mapping, and node labels whenever accelerator hardware types are added.
  3. Require a compatibility matrix and acceptance test spanning BCM, images, schedulers, and host agents before upgrade.
  4. Capture NGC image digests and provenance in versioned job and cluster release manifests.

Explanation

Correct Answer

A:

This preventive practice addresses the underlying failure (Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.) and supports continued assurance through: A forced partition leaves a single active management authority and protects configuration state against conflicting writes.

Incorrect Answers

B:

This prevention control targets another operational risk: A BCM provisioning environment discovers every compute node, yet new GPU hosts join the wrong category and receive CPU-only operating system images. The currently reported mechanism remains: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

C:

This prevention control targets another operational risk: An operations team upgrades BCM directly in production without testing whether existing Slurm and provisioning integrations are compatible with the new management API. The currently reported mechanism remains: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

D:

This prevention control targets another operational risk: Several AI nodes pull a moving NGC container tag at job time, producing different application libraries across otherwise identical distributed training ranks. The currently reported mechanism remains: Redundant managers without proper leadership arbitration can corrupt the provisioning database during network partitions.

NCP-AI Infrastructure (NCP-AII) Practice Tests