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 23

Question 220

During an NVIDIA AI factory acceptance test, engineers discover: A new GPU pod connects all compute racks through one core uplink and cross-rack collectives congest immediately. Which TWO complementary elements are needed: the correct technical action and the strongest independent diagnostic evidence? Select TWO answers.

  1. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.
  2. Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test.
  3. Inspect management API target paths, access scopes, sensor values, chassis actions and independent host inventory.
  4. Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains.
  5. Validate platform TPM provisioning, UEFI secure-boot policy and measured-boot evidence against the approved baseline.

Explanation

Correct Answers

B:

This is the direct operational evidence to investigate: Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test. Observed scenario: A new GPU pod connects all compute racks through one core uplink and cross-rack collectives congest immediately. The governing mechanism is: GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.

D:

This is an appropriate correction for AI factory spine leaf underlay design: Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains. Observed scenario: A new GPU pod connects all compute racks through one core uplink and cross-rack collectives congest immediately. The governing mechanism is: GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. A new GPU pod connects all compute racks through one core uplink and cross-rack collectives congest immediately. GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.

C:

This alternative is relevant to Redfish inventory and IPMI management boundaries but does not directly resolve the current AI factory spine leaf underlay design issue: Inspect management API target paths, access scopes, sensor values, chassis actions and independent host inventory. Observed scenario: A new GPU pod connects all compute racks through one core uplink and cross-rack collectives congest immediately.

E:

This alternative is relevant to TPM state and measured secure boot but does not directly resolve the current AI factory spine leaf underlay design issue: Validate platform TPM provisioning, UEFI secure-boot policy and measured-boot evidence against the approved baseline. Observed scenario: A new GPU pod connects all compute racks through one core uplink and cross-rack collectives congest immediately.

Question 221

A production GPU cluster review identifies this condition: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant. Which operational response is MOST appropriate to address the root issue? Select ONE best answer.

  1. Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains.
  2. Use approved authenticated BMC inventory endpoints, distinguish read-only calls from power actions and corroborate returned component state.
  3. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.
  4. Restore planned airflow direction, containment and intake clearance, and validate inlet thermal conditions at loaded nodes.

Explanation

Correct Answer

A:

This is an appropriate correction for AI factory spine leaf underlay design: Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant.

Incorrect Answers

B:

This alternative is relevant to Redfish inventory and IPMI management boundaries but does not directly resolve the current AI factory spine leaf underlay design issue: Use approved authenticated BMC inventory endpoints, distinguish read-only calls from power actions and corroborate returned component state. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant.

C:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant.

D:

That option concerns Cold aisle inlet air and thermal envelope. GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.

Question 222

In a professional AI infrastructure deployment, a technician observes: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant. Which underlying infrastructure principle BEST explains the issue? Select ONE best answer.

  1. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.
  2. Grounding and electrostatic protection limit personnel hazards and component damage; successful boot alone does not establish safe installation.
  3. GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.
  4. Liquid cooling integrity and appropriate flow are part of accelerator availability and hardware safety, not optional performance tuning.

Explanation

Correct Answer

C:

This is the governing infrastructure principle: GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant. GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.

B:

That option concerns Rack bonding grounding and ESD handling. A direct check is: Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test.

D:

That option concerns Liquid cooling supply and leak interlocks. A direct check is: Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test.

Question 223

The cluster operations team needs to evaluate this situation: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant. Which diagnostic evidence should be obtained FIRST to substantiate the likely fault or configuration gap? Select ONE best answer.

  1. Compare platform power specifications, measured peak draw, PDU branch telemetry, breaker ratings and planned concurrency.
  2. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.
  3. Inspect management API target paths, access scopes, sensor values, chassis actions and independent host inventory.
  4. Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test.

Explanation

Correct Answer

D:

This is the direct operational evidence to investigate: Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant.

Incorrect Answers

A:

That option concerns GPU node peak power budgeting. GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.

B:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant. The governing mechanism is: GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile. The proportionate action is: Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains.

C:

This alternative is relevant to Redfish inventory and IPMI management boundaries but does not directly resolve the current AI factory spine leaf underlay design issue: Inspect management API target paths, access scopes, sensor values, chassis actions and independent host inventory. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant.

Question 224

At a multi-node AI infrastructure handoff, the team records: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant. After corrective work, which observable result BEST confirms that the requirement is met? Select ONE best answer.

  1. All nodes are securely mounted in compatible rack positions and mechanical inspection passes without blocked service access.
  2. Training traffic has the intended aggregate capacity and remains reachable across a permitted single fabric fault.
  3. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.
  4. A simulated cooling fault raises the intended alert and safely reduces or stops workload before damage occurs.

Explanation

Correct Answer

B:

This is an observable verified result, not merely a claimed configuration change: Training traffic has the intended aggregate capacity and remains reachable across a permitted single fabric fault. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant.

Incorrect Answers

A:

That option concerns DGX rack load and rail placement. GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.

C:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant. The governing mechanism is: GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile. The proportionate action is: Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains. A direct check is: Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test. The acceptance condition is: Training traffic has the intended aggregate capacity and remains reachable across a permitted single fabric fault.

D:

This alternative is relevant to Liquid cooling supply and leak interlocks but does not directly resolve the current AI factory spine leaf underlay design issue: A simulated cooling fault raises the intended alert and safely reduces or stops workload before damage occurs. Observed scenario: An expansion plan wires every leaf to only one spine while declaring the fabric fully redundant.

Question 225

During an NVIDIA AI factory acceptance test, engineers discover: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths. Which operational response is MOST appropriate to address the root issue? Select ONE best answer.

  1. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.
  2. Trace each power supply to independent approved rack feeds and test safe loss of either feed within the supported power budget.
  3. Provision unique BMC addresses on protected out-of-band management, rotate initial credentials and test access separately from host networking.
  4. Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains.

Explanation

Correct Answer

D:

This is an appropriate correction for AI factory spine leaf underlay design: Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

B:

This alternative is relevant to Dual-feed rack power and PDU mapping but does not directly resolve the current AI factory spine leaf underlay design issue: Trace each power supply to independent approved rack feeds and test safe loss of either feed within the supported power budget. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

C:

This alternative is relevant to BMC network bring-up and out-of-band separation but does not directly resolve the current AI factory spine leaf underlay design issue: Provision unique BMC addresses on protected out-of-band management, rotate initial credentials and test access separately from host networking. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

Question 226

A production GPU cluster review identifies this condition: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths. Which underlying infrastructure principle BEST explains the issue? Select ONE best answer.

  1. GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.
  2. AI compute power demand varies with accelerator utilization and can exceed idle or average measurements during parallel jobs.
  3. A BMC provides hardware management independent of the host operating system and warrants isolated credential-controlled access.
  4. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.

Explanation

Correct Answer

A:

This is the governing infrastructure principle: GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

Incorrect Answers

B:

This alternative is relevant to GPU node peak power budgeting but does not directly resolve the current AI factory spine leaf underlay design issue: AI compute power demand varies with accelerator utilization and can exceed idle or average measurements during parallel jobs. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

C:

That option concerns BMC network bring-up and out-of-band separation. A direct check is: Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test.

D:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths. GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile.

Question 227

In a professional AI infrastructure deployment, a technician observes: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths. Which diagnostic evidence should be obtained FIRST to substantiate the likely fault or configuration gap? Select ONE best answer.

  1. Inspect management API target paths, access scopes, sensor values, chassis actions and independent host inventory.
  2. Inspect TPM ownership and event logs, secure-boot state, signed firmware path and expected PCR measurements.
  3. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.
  4. Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test.

Explanation

Correct Answer

D:

This is the direct operational evidence to investigate: Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

Incorrect Answers

A:

This alternative is relevant to Redfish inventory and IPMI management boundaries but does not directly resolve the current AI factory spine leaf underlay design issue: Inspect management API target paths, access scopes, sensor values, chassis actions and independent host inventory. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

B:

This alternative is relevant to TPM state and measured secure boot but does not directly resolve the current AI factory spine leaf underlay design issue: Inspect TPM ownership and event logs, secure-boot state, signed firmware path and expected PCR measurements. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

C:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths. The governing mechanism is: GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile. The proportionate action is: Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains.

Question 228

The cluster operations team needs to evaluate this situation: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths. After corrective work, which observable result BEST confirms that the requirement is met? Select ONE best answer.

  1. Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf.
  2. Authorized operators can manage an offline host through the protected BMC while unauthorized network access is denied.
  3. Electrical inspection approves the rack and sensitive boards are installed without ESD or grounding exceptions.
  4. Training traffic has the intended aggregate capacity and remains reachable across a permitted single fabric fault.

Explanation

Correct Answer

D:

This is an observable verified result, not merely a claimed configuration change: Training traffic has the intended aggregate capacity and remains reachable across a permitted single fabric fault. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for AI factory spine leaf underlay design: Assume a single 400G uplink guarantees 400G simultaneous service to every GPU server behind the same leaf. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths. The governing mechanism is: GPU collectives generate heavy east-west traffic; a physically connected fabric may still be oversubscribed or fragile. The proportionate action is: Use a scalable supported leaf-spine topology with adequate underlay links, routing and tested failure domains. A direct check is: Inspect topology diagrams, leaf-spine link maps, ECMP next hops, utilization and a planned spine-failure test. The acceptance condition is: Training traffic has the intended aggregate capacity and remains reachable across a permitted single fabric fault.

B:

This alternative is relevant to BMC network bring-up and out-of-band separation but does not directly resolve the current AI factory spine leaf underlay design issue: Authorized operators can manage an offline host through the protected BMC while unauthorized network access is denied. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

C:

This alternative is relevant to Rack bonding grounding and ESD handling but does not directly resolve the current AI factory spine leaf underlay design issue: Electrical inspection approves the rack and sensitive boards are installed without ESD or grounding exceptions. Observed scenario: A large training cluster is placed on a flat access tree without reviewing east-west bandwidth or failure paths.

Question 229

At a multi-node AI infrastructure handoff, the team records: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links. Which operational response is MOST appropriate to address the root issue? Select ONE best answer.

  1. Declare any InfiniBand leaf-spine network nonblocking solely from a single adapter's line-rate specification.
  2. Check supported CDU supply, flow, leak detection and interlock behavior, then demonstrate controlled thermal response before production.
  3. Validate platform TPM provisioning, UEFI secure-boot policy and measured-boot evidence against the approved baseline.
  4. Validate fabric topology, end-to-end capacity, routing and oversubscription against representative collective communication patterns.

Explanation

Correct Answer

D:

This is an appropriate correction for InfiniBand fat-tree and bisection planning: Validate fabric topology, end-to-end capacity, routing and oversubscription against representative collective communication patterns. Observed scenario: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for InfiniBand fat-tree and bisection planning: Declare any InfiniBand leaf-spine network nonblocking solely from a single adapter's line-rate specification.

B:

This alternative is relevant to Liquid cooling supply and leak interlocks but does not directly resolve the current InfiniBand fat-tree and bisection planning issue: Check supported CDU supply, flow, leak detection and interlock behavior, then demonstrate controlled thermal response before production. Observed scenario: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links.

C:

This alternative is relevant to TPM state and measured secure boot but does not directly resolve the current InfiniBand fat-tree and bisection planning issue: Validate platform TPM provisioning, UEFI secure-boot policy and measured-boot evidence against the approved baseline. Observed scenario: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links.

Question 230

During an NVIDIA AI factory acceptance test, engineers discover: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links. Which TWO complementary elements are needed: the correct technical action and the strongest independent diagnostic evidence? Select TWO answers.

  1. Declare any InfiniBand leaf-spine network nonblocking solely from a single adapter's line-rate specification.
  2. Compare planned leaf/spine port ratios, route distribution, switch counters and all-to-all benchmark results.
  3. Check supported CDU supply, flow, leak detection and interlock behavior, then demonstrate controlled thermal response before production.
  4. Validate fabric topology, end-to-end capacity, routing and oversubscription against representative collective communication patterns.
  5. Inspect TPM ownership and event logs, secure-boot state, signed firmware path and expected PCR measurements.

Explanation

Correct Answers

B:

This is the direct operational evidence to investigate: Compare planned leaf/spine port ratios, route distribution, switch counters and all-to-all benchmark results. Observed scenario: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links.

D:

This is an appropriate correction for InfiniBand fat-tree and bisection planning: Validate fabric topology, end-to-end capacity, routing and oversubscription against representative collective communication patterns. Observed scenario: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links. The governing mechanism is: InfiniBand endpoint link speed does not establish aggregate bisection bandwidth or acceptable fabric path diversity.

Incorrect Answers

A:

This tempting but unsafe shortcut is not a sufficient remediation for InfiniBand fat-tree and bisection planning: Declare any InfiniBand leaf-spine network nonblocking solely from a single adapter's line-rate specification. InfiniBand endpoint link speed does not establish aggregate bisection bandwidth or acceptable fabric path diversity.

C:

This alternative is relevant to Liquid cooling supply and leak interlocks but does not directly resolve the current InfiniBand fat-tree and bisection planning issue: Check supported CDU supply, flow, leak detection and interlock behavior, then demonstrate controlled thermal response before production. Observed scenario: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links. The governing mechanism is: InfiniBand endpoint link speed does not establish aggregate bisection bandwidth or acceptable fabric path diversity.

E:

This alternative is relevant to TPM state and measured secure boot but does not directly resolve the current InfiniBand fat-tree and bisection planning issue: Inspect TPM ownership and event logs, secure-boot state, signed firmware path and expected PCR measurements. Observed scenario: A compute partition launches all-to-all operations but its InfiniBand topology has insufficient inter-tier links.

NCP-AI Infrastructure (NCP-AII) Practice Tests