NCA - AI Infrastructure and Operations NCA-AIIO Practice Test 69
Question 680
A data-center engineer must resolve the following situation: An inference container requests a GPU but can access every accelerator and shared host filesystem on its node. What is the MOST appropriate response?
- Enforce supported device allocation, tenant permissions and container isolation for GPU workloads.
- Standardize supported host driver, container toolkit and GPU Operator lifecycle on eligible worker pools.
- Cordon or drain the node through approved scheduler workflows and coordinate checkpointable tasks.
- Use hardware feature discovery and node labels or resource constraints for workload-specific placement.
Explanation
Correct Answer
A:The appropriate response is to enforce supported device allocation, tenant permissions and container isolation for GPU workloads. This is specific to the case where an inference container requests a GPU but can access every accelerator and shared host filesystem on its node. The mechanism is that scheduler resource claims and runtime security controls jointly determine workload isolation. If ignored, one tenant may access unassigned accelerator capacity or sensitive shared data. Validate the choice using this targeted evidence: A negative access test confirms the workload sees only approved GPUs and mounted storage.
Incorrect Answers
B:This alternative addresses a different requirement: GPU container images work on one host but fail on new workers because drivers and runtime integration differ. It would be appropriate there to standardize supported host driver, container toolkit and GPU Operator lifecycle on eligible worker pools. But the present scenario is about an inference container requests a GPU but can access every accelerator and shared host filesystem on its node. For it, the needed control is to enforce supported device allocation, tenant permissions and container isolation for GPU workloads.
C:This alternative addresses a different requirement: A GPU node is marked for maintenance while the scheduler continues assigning new long-running training jobs to it. It would be appropriate there to cordon or drain the node through approved scheduler workflows and coordinate checkpointable tasks. But the present scenario is about an inference container requests a GPU but can access every accelerator and shared host filesystem on its node. For it, the needed control is to enforce supported device allocation, tenant permissions and container isolation for GPU workloads.
D:This alternative addresses a different requirement: Kubernetes workers contain different GPU types but all nodes are treated as equivalent for AI job placement. It would be appropriate there to use hardware feature discovery and node labels or resource constraints for workload-specific placement. But the present scenario is about an inference container requests a GPU but can access every accelerator and shared host filesystem on its node. For it, the needed control is to enforce supported device allocation, tenant permissions and container isolation for GPU workloads.
Question 681
The design team faces this case: A team chooses an accelerator based only on peak compute rating although the model nearly fills GPU memory. It adopts: Compare usable memory capacity, bandwidth and software support alongside compute performance. What evidence BEST validates the intended outcome?
- A load run records peak allocated memory, bandwidth utilization and sustained task throughput.
- A profile shows sequential coordination remains CPU-resident while parallel compute is offloaded appropriately.
- A trace records transfer and kernel time separately and demonstrates lower total task latency.
- An equivalent matrix workload is compared for numerical correctness and throughput across CPU and GPU.
Explanation
Correct Answer
A:The strongest proof is: A load run records peak allocated memory, bandwidth utilization and sustained task throughput. It directly tests the safeguard to compare usable memory capacity, bandwidth and software support alongside compute performance. The control addresses this specific context: A team chooses an accelerator based only on peak compute rating although the model nearly fills GPU memory. It matters because the fastest arithmetic unit is insufficient when the model and runtime state cannot fit in memory. Without it, the model can fail allocation or thrash despite the accelerator's high advertised FLOPS.
Incorrect Answers
B:This observation would validate a different control: Keep control-flow-heavy sequential orchestration on CPUs and map suitable parallel kernels to GPUs. It fits the case where a data-center technician assumes one GPU is a universal replacement for a CPU in every application. It cannot prove success here; instead the required evidence is a load run records peak allocated memory, bandwidth utilization and sustained task throughput.
C:This observation would validate a different control: Batch transfers and minimize host-device movement where data dependence and memory budgets permit. It fits the case where a short GPU kernel performs little arithmetic but repeatedly copies tiny host buffers across PCIe. It cannot prove success here; instead the required evidence is a load run records peak allocated memory, bandwidth utilization and sustained task throughput.
D:This observation would validate a different control: Select GPUs for appropriate data-parallel arithmetic while retaining CPU responsibility for orchestration. It fits the case where a team ports a highly parallel matrix-multiplication workload to a CPU-only fleet despite abundant accelerator capacity. It cannot prove success here; instead the required evidence is a load run records peak allocated memory, bandwidth utilization and sustained task throughput.
Question 682
A data-center engineer must resolve the following situation: A cluster sees intermittent network errors but its DPU ports and offload services are missing from health dashboards. What is the MOST appropriate response?
- Assess supported DPU offload for suitable networking and storage functions and measure actual net benefit.
- Assign appropriate network, security or storage infrastructure tasks to the DPU rather than model tensor computation.
- Define supported DPU management, update and rollback procedures separate from host OS changes.
- Monitor supported BlueField interfaces, service state and relevant packet or error counters.
Explanation
Correct Answer
D:The appropriate response is to monitor supported BlueField interfaces, service state and relevant packet or error counters. This is specific to the case where a cluster sees intermittent network errors but its DPU ports and offload services are missing from health dashboards. The mechanism is that dPU deployment adds operational elements that need observability alongside GPU and host status. If ignored, invisible offload failures may be mistaken for model or GPU faults and prolong repair. Validate the choice using this targeted evidence: An injected supported offload fault is localized to the DPU service or interface in cluster telemetry.
Incorrect Answers
A:This alternative addresses a different requirement: A busy accelerated server spends significant host CPU cycles processing storage and network control tasks. It would be appropriate there to assess supported DPU offload for suitable networking and storage functions and measure actual net benefit. But the present scenario is about a cluster sees intermittent network errors but its DPU ports and offload services are missing from health dashboards. For it, the needed control is to monitor supported BlueField interfaces, service state and relevant packet or error counters.
B:This alternative addresses a different requirement: A platform team installs a DPU and assumes it will automatically accelerate every CUDA compute kernel. It would be appropriate there to assign appropriate network, security or storage infrastructure tasks to the DPU rather than model tensor computation. But the present scenario is about a cluster sees intermittent network errors but its DPU ports and offload services are missing from health dashboards. For it, the needed control is to monitor supported BlueField interfaces, service state and relevant packet or error counters.
C:This alternative addresses a different requirement: An administrator configures BlueField networking without documenting ownership of DPU software and firmware updates. It would be appropriate there to define supported DPU management, update and rollback procedures separate from host OS changes. But the present scenario is about a cluster sees intermittent network errors but its DPU ports and offload services are missing from health dashboards. For it, the needed control is to monitor supported BlueField interfaces, service state and relevant packet or error counters.
Question 683
The team recommends this safeguard: Use high-fidelity simulation or synthetic data generation with realism and generalization testing. It applies because an autonomous vehicle company trains a vision policy and needs rare-road-condition examples without public-road risk. Which concept justifies the design?
- Simulation can expose AI systems to hazardous edge cases under controlled repeatable conditions.
- Financial anomaly screening needs measured error trade-offs and controlled investigation decisions.
- Image segmentation predicts spatial regions, supporting detailed inspection rather than replacing clinical judgment.
- Grounded generative assistants should tie factual claims to approved content rather than rely only on model memory.
Explanation
Correct Answer
A:The governing principle is that simulation can expose AI systems to hazardous edge cases under controlled repeatable conditions. It supports the measure to use high-fidelity simulation or synthetic data generation with realism and generalization testing. because an autonomous vehicle company trains a vision policy and needs rare-road-condition examples without public-road risk. Without the control, a model may look robust in synthetic scenes but fail on real-world visual distributions. The most useful acceptance evidence is: The validation plan measures scenario coverage and compares simulator and real-world held-out performance.
Incorrect Answers
B:That principle is relevant to this distinct case: A financial institution wants to flag unusual transaction patterns with tight limits on missed fraud and false alarms. The associated response would be to select a fraud-scoring classifier with monitored precision, recall, drift and human escalation rules. It does not describe the central reason for this case, which is that simulation can expose AI systems to hazardous edge cases under controlled repeatable conditions.
C:That principle is relevant to this distinct case: A radiology team needs to highlight suspicious structures in diagnostic images for qualified human review. The associated response would be to evaluate computer-vision segmentation with clinically governed validation and human oversight. It does not describe the central reason for this case, which is that simulation can expose AI systems to hazardous edge cases under controlled repeatable conditions.
D:That principle is relevant to this distinct case: A customer-service team wants an LLM to draft responses using approved knowledge-base articles. The associated response would be to combine retrieval-grounded generation with source attribution and a human escalation policy. It does not describe the central reason for this case, which is that simulation can expose AI systems to hazardous edge cases under controlled repeatable conditions.
Question 684
An organization is evaluating an AI deployment: Kubernetes workers contain different GPU types but all nodes are treated as equivalent for AI job placement. What solution MOST directly fits the stated need?
- Use hardware feature discovery and node labels or resource constraints for workload-specific placement.
- Cordon or drain the node through approved scheduler workflows and coordinate checkpointable tasks.
- Install and validate supported GPU drivers, NVIDIA device integration and resource advertisement.
- Define resource quotas, scheduling priorities and fair-share policy with accountable ownership.
Explanation
Correct Answer
A:The appropriate response is to use hardware feature discovery and node labels or resource constraints for workload-specific placement. This is specific to the case where kubernetes workers contain different GPU types but all nodes are treated as equivalent for AI job placement. The mechanism is that schedulers need trustworthy accelerator capability metadata to place workloads on compatible devices. If ignored, a model needing a specific GPU feature could be sent to an incompatible node. Validate the choice using this targeted evidence: A scheduling test places the constrained job on a qualifying node and rejects unsupported nodes.
Incorrect Answers
B:This alternative addresses a different requirement: A GPU node is marked for maintenance while the scheduler continues assigning new long-running training jobs to it. It would be appropriate there to cordon or drain the node through approved scheduler workflows and coordinate checkpointable tasks. But the present scenario is about kubernetes workers contain different GPU types but all nodes are treated as equivalent for AI job placement. For it, the needed control is to use hardware feature discovery and node labels or resource constraints for workload-specific placement.
C:This alternative addresses a different requirement: A Kubernetes AI cluster contains NVIDIA GPUs but pod resource requests are invisible to the control plane. It would be appropriate there to install and validate supported GPU drivers, NVIDIA device integration and resource advertisement. But the present scenario is about kubernetes workers contain different GPU types but all nodes are treated as equivalent for AI job placement. For it, the needed control is to use hardware feature discovery and node labels or resource constraints for workload-specific placement.
D:This alternative addresses a different requirement: Multiple research teams compete for a shared GPU fleet and one account submits hundreds of low-priority jobs. It would be appropriate there to define resource quotas, scheduling priorities and fair-share policy with accountable ownership. But the present scenario is about kubernetes workers contain different GPU types but all nodes are treated as equivalent for AI job placement. For it, the needed control is to use hardware feature discovery and node labels or resource constraints for workload-specific placement.
Question 685
An AI infrastructure team faces this requirement: A server has eight accelerators but the design incorrectly assumes they all communicate through direct PCIe peer paths. Which action BEST addresses the operational constraint?
- Document the actual NVSwitch or NVLink GPU topology and select supported collective patterns.
- Identify and protect shared control-plane and storage failure domains with appropriate redundancy.
- Separate management or login responsibilities from dedicated GPU compute workers.
- Provide shared high-throughput storage and a protected checkpoint repository sized for concurrent jobs.
Explanation
Correct Answer
A:The appropriate response is to document the actual NVSwitch or NVLink GPU topology and select supported collective patterns. This is specific to the case where a server has eight accelerators but the design incorrectly assumes they all communicate through direct PCIe peer paths. The mechanism is that gPU interconnect topology determines efficient communication routes inside an accelerated node. If ignored, an incorrect topology assumption can make model-parallel communication slower than predicted. Validate the choice using this targeted evidence: Topology discovery and benchmark results match the documented high-bandwidth inter-GPU paths.
Incorrect Answers
B:This alternative addresses a different requirement: A cluster stores all training metadata and management state on one server without recovery copies. It would be appropriate there to identify and protect shared control-plane and storage failure domains with appropriate redundancy. But the present scenario is about a server has eight accelerators but the design incorrectly assumes they all communicate through direct PCIe peer paths. For it, the needed control is to document the actual NVSwitch or NVLink GPU topology and select supported collective patterns.
C:This alternative addresses a different requirement: Researchers run heavy model training directly on the only cluster login node used for submissions and admin work. It would be appropriate there to separate management or login responsibilities from dedicated GPU compute workers. But the present scenario is about a server has eight accelerators but the design incorrectly assumes they all communicate through direct PCIe peer paths. For it, the needed control is to document the actual NVSwitch or NVLink GPU topology and select supported collective patterns.
D:This alternative addresses a different requirement: A new AI cluster relies on local workstation disks for shared multi-node training data and model checkpoints. It would be appropriate there to provide shared high-throughput storage and a protected checkpoint repository sized for concurrent jobs. But the present scenario is about a server has eight accelerators but the design incorrectly assumes they all communicate through direct PCIe peer paths. For it, the needed control is to document the actual NVSwitch or NVLink GPU topology and select supported collective patterns.
Question 686
A data-center manager declines to adopt a supported cuDNN-enabled framework and validate the convolution workload with representative input shapes. The operational context is: A computer vision team has CUDA kernels but repeatedly rewrites convolution primitives while optimizing its neural network training code. Which risk is MOST direct?
- Handcrafted wrappers may create inconsistent deployment controls or mismatched supported model dependencies.
- Training may consume excessive engineering time or underuse GPU compute despite otherwise correct model math.
- The preparation stage can dominate wall-clock time even when model training itself is highly accelerated.
- Model deployments can become operationally fragile and fail to use available GPU capacity effectively.
Explanation
Correct Answer
B:The primary exposure is that training may consume excessive engineering time or underuse GPU compute despite otherwise correct model math. This follows from a computer vision team has CUDA kernels but repeatedly rewrites convolution primitives while optimizing its neural network training code. where the omitted measure was to adopt a supported cuDNN-enabled framework and validate the convolution workload with representative input shapes. The technical reason is that cuDNN supplies GPU-optimized deep-learning primitives so application teams need not hand-code every operator. A good operational validation after repair would be: A controlled benchmark completes the intended convolution operations with correct outputs and measured accelerator utilization.
Incorrect Answers
A:This exposure arises in a different situation: A support team wants to deploy a packaged language-model microservice without maintaining custom inference server wrappers. There, the corresponding safeguard is to select a supported NVIDIA NIM microservice and verify its model, runtime and deployment prerequisites. In the present case, the more direct risk is that training may consume excessive engineering time or underuse GPU compute despite otherwise correct model math.
C:This exposure arises in a different situation: An analytics department runs large tabular transformations on CPUs while its GPUs remain idle between training jobs. There, the corresponding safeguard is to assess RAPIDS GPU-accelerated data processing where the data format and operators are compatible. In the present case, the more direct risk is that training may consume excessive engineering time or underuse GPU compute despite otherwise correct model math.
D:This exposure arises in a different situation: An organization manually writes a different HTTP serving process for each trained model, making batching and health reporting inconsistent. There, the corresponding safeguard is to evaluate NVIDIA Triton Inference Server with suitable model repositories, batching and protocol endpoints. In the present case, the more direct risk is that training may consume excessive engineering time or underuse GPU compute despite otherwise correct model math.
Question 687
A team omits this measure: Include checkpoint size, frequency, retention and restore throughput in capacity and storage planning. In this environment: A training program uses frequent multi-terabyte checkpoints but reserves storage for only its original dataset. Which resulting problem is MOST likely?
- A full storage volume can interrupt runs or prevent recovery after a node failure.
- A procurement plan may understate memory by several multiples and stall the training job.
- GPU resources remain expensive and idle when a storage system cannot supply enough samples.
- The team may procure overspecified GPUs while leaving a storage or CPU bottleneck unresolved.
Explanation
Correct Answer
A:The primary exposure is that a full storage volume can interrupt runs or prevent recovery after a node failure. This follows from a training program uses frequent multi-terabyte checkpoints but reserves storage for only its original dataset. where the omitted measure was to include checkpoint size, frequency, retention and restore throughput in capacity and storage planning. The technical reason is that reliable training infrastructure must store and restore evolving state as well as original training data. A good operational validation after repair would be: A checkpoint-and-restore exercise completes inside recovery targets with sufficient storage headroom.
Incorrect Answers
B:This exposure arises in a different situation: A team estimates LLM training memory from the file size of weights and omits optimizer and gradient storage. There, the corresponding safeguard is to model total training-state memory and evaluate supported sharding or memory-saving strategies. In the present case, the more direct risk is that a full storage volume can interrupt runs or prevent recovery after a node failure.
C:This exposure arises in a different situation: A vision training job reads millions of small images from one slow network file share and leaves accelerators waiting. There, the corresponding safeguard is to plan storage throughput, metadata operations and prefetch behavior using representative input data. In the present case, the more direct risk is that a full storage volume can interrupt runs or prevent recovery after a node failure.
D:This exposure arises in a different situation: A platform engineer sizes a new cluster from vendor peak GPU FLOPS without profiling the actual training pipeline. There, the corresponding safeguard is to benchmark representative iteration time, memory use and input-data rate before finalizing hardware sizing. In the present case, the more direct risk is that a full storage volume can interrupt runs or prevent recovery after a node failure.
Question 688
The design team faces this case: A team scales training to many GPU nodes but iteration time rises because gradient exchanges dominate compute. It adopts: Profile collective communication and adjust topology, batch strategy or network capacity for data-parallel training. What evidence BEST validates the intended outcome?
- The sharded training run completes one correct optimization step with measured communication overhead.
- A scaling test records useful samples per second and all-reduce time at each cluster size.
- A benchmark plots throughput versus p95 latency and selects a batch policy meeting both limits.
- After a controlled interruption, resumed training starts from a checkpoint with consistent progress and metrics.
Explanation
Correct Answer
B:The strongest proof is: A scaling test records useful samples per second and all-reduce time at each cluster size. It directly tests the safeguard to profile collective communication and adjust topology, batch strategy or network capacity for data-parallel training. The control addresses this specific context: A team scales training to many GPU nodes but iteration time rises because gradient exchanges dominate compute. It matters because synchronized data-parallel training exchanges gradients, so collective efficiency influences scaling gains. Without it, adding GPUs may increase cost without proportionally improving completed training work.
Incorrect Answers
A:This observation would validate a different control: Evaluate model parallelism with appropriate tensor or pipeline partitioning and high-bandwidth interconnect needs. It fits the case where a foundation-model training plan exceeds the memory of a single accelerator even after supported precision reductions. It cannot prove success here; instead the required evidence is a scaling test records useful samples per second and all-reduce time at each cluster size.
C:This observation would validate a different control: Evaluate dynamic batching within the permitted queueing delay and model-specific latency budget. It fits the case where an inference service receives many short requests but operates each GPU one request at a time with poor utilization. It cannot prove success here; instead the required evidence is a scaling test records useful samples per second and all-reduce time at each cluster size.
D:This observation would validate a different control: Write versioned training checkpoints to resilient storage and test restart from a known consistent state. It fits the case where a multi-hour distributed model training run loses all progress when one compute node is recycled for maintenance. It cannot prove success here; instead the required evidence is a scaling test records useful samples per second and all-reduce time at each cluster size.
Question 689
In this situation, a recommendation project improves after collecting representative interaction logs but management attributes all gains to GPU upgrades. The team does not evaluate data quantity, label quality and dataset representativeness alongside accelerated hardware improvements. Which consequence BEST characterizes the gap?
- A workload may be described as software-only even when it cannot complete within available CPU time.
- Starting every model from scratch may waste scarce labeled data and long training windows.
- Teams may spend capacity replicating common primitives rather than validating their actual use cases.
- Hardware purchases may deliver little benefit if the data remain biased or insufficient for the target users.
Explanation
Correct Answer
D:The primary exposure is that hardware purchases may deliver little benefit if the data remain biased or insufficient for the target users. This follows from a recommendation project improves after collecting representative interaction logs but management attributes all gains to GPU upgrades. where the omitted measure was to evaluate data quantity, label quality and dataset representativeness alongside accelerated hardware improvements. The technical reason is that recent AI improvements depend on useful training data as well as computing resources and algorithms. A good operational validation after repair would be: A controlled model comparison isolates the impact of data improvements from compute changes.
Incorrect Answers
A:This exposure arises in a different situation: A research group trains larger neural networks after acquiring accelerators with much higher parallel throughput. There, the corresponding safeguard is to identify accessible accelerated computation as an enabler of feasible large-scale model training. In the present case, the more direct risk is that hardware purchases may deliver little benefit if the data remain biased or insufficient for the target users.
B:This exposure arises in a different situation: An applied AI team starts from an existing pretrained vision model rather than training its representation from random weights. There, the corresponding safeguard is to use transfer learning appropriately and validate downstream quality with the domain-specific dataset. In the present case, the more direct risk is that hardware purchases may deliver little benefit if the data remain biased or insufficient for the target users.
C:This exposure arises in a different situation: Two groups can now build the same neural architecture using mature open frameworks instead of custom kernels. There, the corresponding safeguard is to recognize the role of widely available frameworks and reusable libraries in expanding AI adoption. In the present case, the more direct risk is that hardware purchases may deliver little benefit if the data remain biased or insufficient for the target users.
Question 690
An organization is evaluating an AI deployment: An accelerated server specification lists GPUs and CPUs but fails to allocate suitable network adapters or high-bandwidth storage. What solution MOST directly fits the stated need?
- Identify and protect shared control-plane and storage failure domains with appropriate redundancy.
- Design logical or physical management, storage and workload network separation appropriate to risk.
- Map CPU, GPU, NIC, storage and interconnect components to the complete data path and workload profile.
- Provide shared high-throughput storage and a protected checkpoint repository sized for concurrent jobs.
Explanation
Correct Answer
C:The appropriate response is to map CPU, GPU, NIC, storage and interconnect components to the complete data path and workload profile. This is specific to the case where an accelerated server specification lists GPUs and CPUs but fails to allocate suitable network adapters or high-bandwidth storage. The mechanism is that aI clusters need coordinated compute, communications and storage components rather than GPU devices alone. If ignored, a high-end accelerator server can be idle while missing network or storage resources constrain useful work. Validate the choice using this targeted evidence: A system diagram and benchmark show that every critical data path meets measured workload requirements.
Incorrect Answers
A:This alternative addresses a different requirement: A cluster stores all training metadata and management state on one server without recovery copies. It would be appropriate there to identify and protect shared control-plane and storage failure domains with appropriate redundancy. But the present scenario is about an accelerated server specification lists GPUs and CPUs but fails to allocate suitable network adapters or high-bandwidth storage. For it, the needed control is to map CPU, GPU, NIC, storage and interconnect components to the complete data path and workload profile.
B:This alternative addresses a different requirement: A research environment shares the same traffic interface for administrative SSH, storage recovery and large all-reduce jobs. It would be appropriate there to design logical or physical management, storage and workload network separation appropriate to risk. But the present scenario is about an accelerated server specification lists GPUs and CPUs but fails to allocate suitable network adapters or high-bandwidth storage. For it, the needed control is to map CPU, GPU, NIC, storage and interconnect components to the complete data path and workload profile.
D:This alternative addresses a different requirement: A new AI cluster relies on local workstation disks for shared multi-node training data and model checkpoints. It would be appropriate there to provide shared high-throughput storage and a protected checkpoint repository sized for concurrent jobs. But the present scenario is about an accelerated server specification lists GPUs and CPUs but fails to allocate suitable network adapters or high-bandwidth storage. For it, the needed control is to map CPU, GPU, NIC, storage and interconnect components to the complete data path and workload profile.