NCA - AI Infrastructure and Operations NCA-AIIO Practice Test 19
Question 180
The selected AI infrastructure safeguard is not implemented: Distinguish generative modeling from discriminative classification when selecting model evaluation criteria. The observed situation is: A text model generates new summaries while a separate classifier assigns each message to a fixed department. What exposure remains?
- Testing only category accuracy can miss hallucinated or unsafe content from the generative model.
- The project may use unsuitable classification metrics and overlook unsafe exploration behavior.
- Operators can overlook model drift while assuming a deployed checkpoint adapts automatically.
- The project may promise improvements from additional data even though its rule engine never learns.
Explanation
Correct Answer
A:The primary exposure is that testing only category accuracy can miss hallucinated or unsafe content from the generative model. This follows from a text model generates new summaries while a separate classifier assigns each message to a fixed department. where the omitted measure was to distinguish generative modeling from discriminative classification when selecting model evaluation criteria. The technical reason is that generative systems create content or sequences while classifiers predict categories for observed inputs. A good operational validation after repair would be: The evaluation suite scores factual support of summaries separately from the department classification accuracy.
Incorrect Answers
B:This exposure arises in a different situation: A robot learns control actions by interacting with a simulated environment and receiving delayed reward signals. There, the corresponding safeguard is to recognize reinforcement learning and define rewards, actions and safe simulation boundaries. In the present case, the more direct risk is that testing only category accuracy can miss hallucinated or unsafe content from the generative model.
C:This exposure arises in a different situation: A company says its production fraud-scoring API is still learning from each request although model weights never change. There, the corresponding safeguard is to separate trained-model inference from explicit retraining or online-learning update processes. In the present case, the more direct risk is that testing only category accuracy can miss hallucinated or unsafe content from the generative model.
D:This exposure arises in a different situation: A project hard-codes hundreds of manually chosen rules to classify support tickets, yet the team describes it as a trained ML model. There, the corresponding safeguard is to distinguish fixed rule-based automation from models that learn patterns from examples during training. In the present case, the more direct risk is that testing only category accuracy can miss hallucinated or unsafe content from the generative model.
Question 181
A technology review accepts this control: Confirm accelerator data-type support and validate precision choice against convergence requirements. Which observed result provides the STRONGEST evidence of effectiveness?
- A topology-aware bandwidth test and training profile show acceptable GPU-to-GPU transfer time.
- A controlled epoch comparison measures convergence and throughput at the selected precision setting.
- A topology-aware test reports lower communication latency after matching GPU, CPU and NIC locality.
- A sizing report attributes measured step time to compute, communication and input-data stages.
Explanation
Correct Answer
B:The strongest proof is: A controlled epoch comparison measures convergence and throughput at the selected precision setting. It directly tests the safeguard to confirm accelerator data-type support and validate precision choice against convergence requirements. The control addresses this specific context: A training team buys accelerators rated for maximum low-precision throughput while requiring strict numeric convergence. It matters because peak accelerator arithmetic specifications depend on eligible operations and supported precision modes. Without it, the planned training may fail numerical acceptance even though advertised compute looks excellent.
Incorrect Answers
A:This observation would validate a different control: Select an appropriate high-bandwidth GPU interconnect topology for the planned multi-GPU communication. It fits the case where a deep-learning training server contains several GPUs but inter-GPU copies must cross a slower host path. It cannot prove success here; instead the required evidence is a controlled epoch comparison measures convergence and throughput at the selected precision setting.
C:This observation would validate a different control: Inspect CPU NUMA, PCIe and GPU-to-NIC affinity before assigning communication-heavy training jobs. It fits the case where two GPUs share a host but one is connected through a constrained PCIe path to the preferred network adapter. It cannot prove success here; instead the required evidence is a controlled epoch comparison measures convergence and throughput at the selected precision setting.
D:This observation would validate a different control: Benchmark representative iteration time, memory use and input-data rate before finalizing hardware sizing. It fits the case where a platform engineer sizes a new cluster from vendor peak GPU FLOPS without profiling the actual training pipeline. It cannot prove success here; instead the required evidence is a controlled epoch comparison measures convergence and throughput at the selected precision setting.
Question 182
An architecture review proposes: Maintain an authoritative inventory linking device identifiers to node, rack, NIC, firmware and support ownership. The underlying requirement is: Engineers cannot determine which network port and GPU belongs to a rack because accelerator inventory uses manual notes. What technical rationale MOST strongly supports it?
- Isolation of control traffic prevents data-plane congestion from blocking administrative response.
- Training storage must support distributed access, ingestion and recoverability across multiple nodes.
- Asset discovery and topology inventory are foundational for maintenance and accelerated workload diagnosis.
- The availability of cluster services depends on management and data services as well as compute nodes.
Explanation
Correct Answer
C:The governing principle is that asset discovery and topology inventory are foundational for maintenance and accelerated workload diagnosis. It supports the measure to maintain an authoritative inventory linking device identifiers to node, rack, NIC, firmware and support ownership. because engineers cannot determine which network port and GPU belongs to a rack because accelerator inventory uses manual notes. Without the control, a failing component may be replaced in the wrong host or its support contract may be missed. The most useful acceptance evidence is: A sampled GPU serial and NIC link map correctly to the physical rack and current firmware inventory.
Incorrect Answers
A:That principle is relevant to this distinct case: A research environment shares the same traffic interface for administrative SSH, storage recovery and large all-reduce jobs. The associated response would be to design logical or physical management, storage and workload network separation appropriate to risk. It does not describe the central reason for this case, which is that asset discovery and topology inventory are foundational for maintenance and accelerated workload diagnosis.
B:That principle is relevant to this distinct case: A new AI cluster relies on local workstation disks for shared multi-node training data and model checkpoints. The associated response would be to provide shared high-throughput storage and a protected checkpoint repository sized for concurrent jobs. It does not describe the central reason for this case, which is that asset discovery and topology inventory are foundational for maintenance and accelerated workload diagnosis.
D:That principle is relevant to this distinct case: A cluster stores all training metadata and management state on one server without recovery copies. The associated response would be to identify and protect shared control-plane and storage failure domains with appropriate redundancy. It does not describe the central reason for this case, which is that asset discovery and topology inventory are foundational for maintenance and accelerated workload diagnosis.
Question 183
The proposed approach is to use supported NVLink or NVSwitch paths for local GPU exchange rather than assuming every copy crosses Ethernet. In the following setting, a scale-up system exchanges large tensors between GPUs inside one chassis and compares that traffic with rack networking. Which principle BEST supports the approach?
- Local GPU interconnects and external rack fabrics serve different scopes of the data path.
- A scalable high-speed fabric requires multiple robust paths and sufficient aggregation capacity.
- High-performance InfiniBand networks can support latency-sensitive collective traffic between AI nodes.
- Modern Ethernet fabrics can support suitable AI networking when loss and congestion are managed.
Explanation
Correct Answer
A:The governing principle is that local GPU interconnects and external rack fabrics serve different scopes of the data path. It supports the measure to use supported NVLink or NVSwitch paths for local GPU exchange rather than assuming every copy crosses Ethernet. because a scale-up system exchanges large tensors between GPUs inside one chassis and compares that traffic with rack networking. Without the control, routing local tensor transfers through a slower external path can waste accelerator time. The most useful acceptance evidence is: Topology mapping confirms intra-node tensors use the expected high-bandwidth GPU interconnect.
Incorrect Answers
B:That principle is relevant to this distinct case: A data center expands the number of training racks but relies on a single aggregation switch for every east-west transfer. The associated response would be to design an appropriate resilient leaf-spine topology and validate required bisection bandwidth. It does not describe the central reason for this case, which is that local GPU interconnects and external rack fabrics serve different scopes of the data path.
C:That principle is relevant to this distinct case: A high-performance computing group trains large models requiring fast node-to-node collective operations. The associated response would be to evaluate a suitably sized InfiniBand fabric with documented latency, bandwidth and management requirements. It does not describe the central reason for this case, which is that local GPU interconnects and external rack fabrics serve different scopes of the data path.
D:That principle is relevant to this distinct case: A cloud provider standardizes on Ethernet switching but wants accelerated AI traffic over the existing operational model. The associated response would be to assess a tuned high-speed Ethernet or RoCE solution rather than assuming ordinary LAN defaults are enough. It does not describe the central reason for this case, which is that local GPU interconnects and external rack fabrics serve different scopes of the data path.
Question 184
An operations team implements: Apply least-privilege administration and protected policy management for DPU-hosted infrastructure controls. Which test result would MOST convincingly show the approach works?
- A responsibilities matrix distinguishes GPU tensor operations from DPU infrastructure offload tasks.
- An injected supported offload fault is localized to the DPU service or interface in cluster telemetry.
- A controlled unauthorized-change attempt is denied and records an auditable policy event.
- Inventory and maintenance records show compatible host and DPU versions and a tested recovery approach.
Explanation
Correct Answer
C:The strongest proof is: A controlled unauthorized-change attempt is denied and records an auditable policy event. It directly tests the safeguard to apply least-privilege administration and protected policy management for DPU-hosted infrastructure controls. The control addresses this specific context: A data-center design uses a DPU for security enforcement but permits every workload administrator to reconfigure its policies. It matters because infrastructure isolation benefits disappear when untrusted tenants can modify offloaded enforcement rules. Without it, a tenant administrator may bypass network policies or alter access outside the assigned workload.
Incorrect Answers
A:This observation would validate a different control: Assign appropriate network, security or storage infrastructure tasks to the DPU rather than model tensor computation. It fits the case where a platform team installs a DPU and assumes it will automatically accelerate every CUDA compute kernel. It cannot prove success here; instead the required evidence is a controlled unauthorized-change attempt is denied and records an auditable policy event.
B:This observation would validate a different control: Monitor supported BlueField interfaces, service state and relevant packet or error counters. It fits the case where a cluster sees intermittent network errors but its DPU ports and offload services are missing from health dashboards. It cannot prove success here; instead the required evidence is a controlled unauthorized-change attempt is denied and records an auditable policy event.
D:This observation would validate a different control: Define supported DPU management, update and rollback procedures separate from host OS changes. It fits the case where an administrator configures BlueField networking without documenting ownership of DPU software and firmware updates. It cannot prove success here; instead the required evidence is a controlled unauthorized-change attempt is denied and records an auditable policy event.
Question 185
A technology review accepts this control: Install and validate supported GPU drivers, NVIDIA device integration and resource advertisement. Which observed result provides the STRONGEST evidence of effectiveness?
- A scheduling test places the constrained job on a qualifying node and rejects unsupported nodes.
- A negative access test confirms the workload sees only approved GPUs and mounted storage.
- An admission test respects tenant quotas while an authorized high-priority job receives agreed capacity.
- An authorized test pod requests one advertised GPU and can execute the approved diagnostic workload.
Explanation
Correct Answer
D:The strongest proof is: An authorized test pod requests one advertised GPU and can execute the approved diagnostic workload. It directly tests the safeguard to install and validate supported GPU drivers, NVIDIA device integration and resource advertisement. The control addresses this specific context: A Kubernetes AI cluster contains NVIDIA GPUs but pod resource requests are invisible to the control plane. It matters because kubernetes requires an appropriate device discovery and allocation path to schedule accelerator workloads. Without it, gPU pods may remain pending or start without usable devices even when hardware exists.
Incorrect Answers
A:This observation would validate a different control: Use hardware feature discovery and node labels or resource constraints for workload-specific placement. It fits the case where kubernetes workers contain different GPU types but all nodes are treated as equivalent for AI job placement. It cannot prove success here; instead the required evidence is an authorized test pod requests one advertised GPU and can execute the approved diagnostic workload.
B:This observation would validate a different control: Enforce supported device allocation, tenant permissions and container isolation for GPU workloads. It fits the case where an inference container requests a GPU but can access every accelerator and shared host filesystem on its node. It cannot prove success here; instead the required evidence is an authorized test pod requests one advertised GPU and can execute the approved diagnostic workload.
C:This observation would validate a different control: Define resource quotas, scheduling priorities and fair-share policy with accountable ownership. It fits the case where multiple research teams compete for a shared GPU fleet and one account submits hundreds of low-priority jobs. It cannot prove success here; instead the required evidence is an authorized test pod requests one advertised GPU and can execute the approved diagnostic workload.
Question 186
The team recommends this safeguard: Keep an appropriately isolated final test set and document evaluation metrics before selecting a release. It applies because a team reports model accuracy using the same data that were repeatedly used for hyperparameter selection. Which concept justifies the design?
- Model deployment requires runtime integration and operational acceptance separate from offline training accuracy.
- Model validation must assess generalization on evidence not repeatedly optimized during model selection.
- AI lifecycles begin with governed data preparation and traceable inputs, not just accelerator allocation.
- Reproducible AI experiments require the parameters and input versions that produced a checkpoint.
Explanation
Correct Answer
B:The governing principle is that model validation must assess generalization on evidence not repeatedly optimized during model selection. It supports the measure to keep an appropriately isolated final test set and document evaluation metrics before selecting a release. because a team reports model accuracy using the same data that were repeatedly used for hyperparameter selection. Without the control, repeated test-set tuning can exaggerate real-world quality and lead to unsafe deployment decisions. The most useful acceptance evidence is: A release report includes held-out test metrics and clearly separated tuning and final evaluation datasets.
Incorrect Answers
A:That principle is relevant to this distinct case: A model-serving change reaches production without checking its API schema, resource limits or startup health. The associated response would be to use staged deployment validation for model artifact, endpoint compatibility, GPU resources and readiness. It does not describe the central reason for this case, which is that model validation must assess generalization on evidence not repeatedly optimized during model selection.
C:That principle is relevant to this distinct case: A model team trains on datasets whose source and consent status are not recorded across experiments. The associated response would be to establish versioned data ingestion, lineage and access controls before model-development experiments. It does not describe the central reason for this case, which is that model validation must assess generalization on evidence not repeatedly optimized during model selection.
D:That principle is relevant to this distinct case: Research experiments change learning rates and seeds without recording configuration or resulting checkpoints. The associated response would be to track hyperparameters, dataset versions, random seeds and artifacts in an experiment registry. It does not describe the central reason for this case, which is that model validation must assess generalization on evidence not repeatedly optimized during model selection.
Question 187
A data-center manager declines to standardize supported host driver, container toolkit and GPU Operator lifecycle on eligible worker pools. The operational context is: GPU container images work on one host but fail on new workers because drivers and runtime integration differ. Which risk is MOST direct?
- Jobs can start without the GPUs they expect or remain indefinitely queued despite apparent free CPUs.
- A scheduling success may still lead to failed CUDA initialization on a mismatched node.
- GPU pods may remain pending or start without usable devices even when hardware exists.
- One tenant can monopolize GPUs and starve urgent inference or research commitments.
Explanation
Correct Answer
B:The primary exposure is that a scheduling success may still lead to failed CUDA initialization on a mismatched node. This follows from gPU container images work on one host but fail on new workers because drivers and runtime integration differ. where the omitted measure was to standardize supported host driver, container toolkit and GPU Operator lifecycle on eligible worker pools. The technical reason is that container orchestration depends on consistent host-side accelerator drivers and device exposure. A good operational validation after repair would be: A fresh worker joins the cluster and runs the same versioned GPU test container successfully.
Incorrect Answers
A:This exposure arises in a different situation: Researchers request four GPUs from Slurm but jobs are assigned to nodes without enough contiguous available devices. There, the corresponding safeguard is to configure scheduler GPU resource accounting and validated device placement constraints. In the present case, the more direct risk is that a scheduling success may still lead to failed CUDA initialization on a mismatched node.
C:This exposure arises in a different situation: A Kubernetes AI cluster contains NVIDIA GPUs but pod resource requests are invisible to the control plane. There, the corresponding safeguard is to install and validate supported GPU drivers, NVIDIA device integration and resource advertisement. In the present case, the more direct risk is that a scheduling success may still lead to failed CUDA initialization on a mismatched node.
D:This exposure arises in a different situation: Multiple research teams compete for a shared GPU fleet and one account submits hundreds of low-priority jobs. There, the corresponding safeguard is to define resource quotas, scheduling priorities and fair-share policy with accountable ownership. In the present case, the more direct risk is that a scheduling success may still lead to failed CUDA initialization on a mismatched node.
Question 188
The proposed approach is to correlate per-GPU compute, memory, scheduling contention and request tail latency. In the following setting, an inference service meets average latency but occasional requests become slow when multiple GPU tenants share the node. Which principle BEST supports the approach?
- GPU occupancy does not reveal all latency introduced by communication outside compute kernels.
- GPU interconnect health directly influences distributed work within supported accelerator systems.
- Xid diagnostics provide context for GPU software or hardware failures and require supported interpretation.
- Averages obscure contention and intermittent device saturation affecting a subset of customers.
Explanation
Correct Answer
D:The governing principle is that averages obscure contention and intermittent device saturation affecting a subset of customers. It supports the measure to correlate per-GPU compute, memory, scheduling contention and request tail latency. because an inference service meets average latency but occasional requests become slow when multiple GPU tenants share the node. Without the control, a shared accelerator may meet aggregate throughput while violating individual service guarantees. The most useful acceptance evidence is: Per-tenant load tests link p95 latency spikes to GPU resource contention and validate an isolation change.
Incorrect Answers
A:That principle is relevant to this distinct case: GPU utilization looks normal but a pipeline spends much of each step copying tensors through a constrained PCIe link. The associated response would be to measure host-device transfer duration, PCIe utilization and data-path topology during the real workload. It does not describe the central reason for this case, which is that averages obscure contention and intermittent device saturation affecting a subset of customers.
B:That principle is relevant to this distinct case: Multi-GPU jobs slow after cabling work and a topology view shows one NVLink path with elevated errors. The associated response would be to inspect supported NVLink status and error counters alongside application collective performance. It does not describe the central reason for this case, which is that averages obscure contention and intermittent device saturation affecting a subset of customers.
C:That principle is relevant to this distinct case: A containerized training job unexpectedly exits and system logs contain GPU Xid error messages. The associated response would be to correlate NVIDIA Xid events with driver status, hardware health and the affected job timeline. It does not describe the central reason for this case, which is that averages obscure contention and intermittent device saturation affecting a subset of customers.
Question 189
A team omits this measure: Evaluate BlueField DPU-supported trusted infrastructure processing and its administrative trust boundaries. In this environment: A security architect wants infrastructure service isolation from host applications that may be compromised. Which resulting problem is MOST likely?
- The project may buy or configure unsuitable hardware and fail to improve neural-network compute speed.
- Untracked DPU versions can create connectivity regressions or block timely security remediation.
- A tenant administrator may bypass network policies or alter access outside the assigned workload.
- A compromised host may gain inappropriate control of shared infrastructure functions if boundaries are absent.
Explanation
Correct Answer
D:The primary exposure is that a compromised host may gain inappropriate control of shared infrastructure functions if boundaries are absent. This follows from a security architect wants infrastructure service isolation from host applications that may be compromised. where the omitted measure was to evaluate BlueField DPU-supported trusted infrastructure processing and its administrative trust boundaries. The technical reason is that dedicated infrastructure processing can separate selected network and security duties from general host code. A good operational validation after repair would be: An access test demonstrates the host cannot alter protected DPU management settings without authorization.
Incorrect Answers
A:This exposure arises in a different situation: A platform team installs a DPU and assumes it will automatically accelerate every CUDA compute kernel. There, the corresponding safeguard is to assign appropriate network, security or storage infrastructure tasks to the DPU rather than model tensor computation. In the present case, the more direct risk is that a compromised host may gain inappropriate control of shared infrastructure functions if boundaries are absent.
B:This exposure arises in a different situation: An administrator configures BlueField networking without documenting ownership of DPU software and firmware updates. There, the corresponding safeguard is to define supported DPU management, update and rollback procedures separate from host OS changes. In the present case, the more direct risk is that a compromised host may gain inappropriate control of shared infrastructure functions if boundaries are absent.
C:This exposure arises in a different situation: A data-center design uses a DPU for security enforcement but permits every workload administrator to reconfigure its policies. There, the corresponding safeguard is to apply least-privilege administration and protected policy management for DPU-hosted infrastructure controls. In the present case, the more direct risk is that a compromised host may gain inappropriate control of shared infrastructure functions if boundaries are absent.
Question 190
An AI infrastructure team faces this requirement: A model endpoint experiences out-of-memory failures only during long prompts and concurrent sessions. Which action BEST addresses the operational constraint?
- Correlate SM utilization with CPU data loading, communication time and memory bandwidth measures.
- Measure host-device transfer duration, PCIe utilization and data-path topology during the real workload.
- Use NVIDIA DCGM monitoring and supported telemetry exporters for centralized device diagnostics.
- Monitor used and available GPU memory across realistic serving contexts and concurrent workload sizes.
Explanation
Correct Answer
D:The appropriate response is to monitor used and available GPU memory across realistic serving contexts and concurrent workload sizes. This is specific to the case where a model endpoint experiences out-of-memory failures only during long prompts and concurrent sessions. The mechanism is that gPU memory telemetry must capture runtime allocations and usage peaks rather than only static model size. If ignored, the service can fail during demand peaks although idle memory looks sufficient. Validate the choice using this targeted evidence: A high-context load test reports peak memory and triggers a controlled alert before allocation failure.
Incorrect Answers
A:This alternative addresses a different requirement: A GPU has low reported utilization while its training job is slow and the team assumes the accelerator is defective. It would be appropriate there to correlate SM utilization with CPU data loading, communication time and memory bandwidth measures. But the present scenario is about a model endpoint experiences out-of-memory failures only during long prompts and concurrent sessions. For it, the needed control is to monitor used and available GPU memory across realistic serving contexts and concurrent workload sizes.
B:This alternative addresses a different requirement: GPU utilization looks normal but a pipeline spends much of each step copying tensors through a constrained PCIe link. It would be appropriate there to measure host-device transfer duration, PCIe utilization and data-path topology during the real workload. But the present scenario is about a model endpoint experiences out-of-memory failures only during long prompts and concurrent sessions. For it, the needed control is to monitor used and available GPU memory across realistic serving contexts and concurrent workload sizes.
C:This alternative addresses a different requirement: A fleet has thousands of accelerators but operators manually run nvidia-smi on each node to identify unhealthy GPUs. It would be appropriate there to use NVIDIA DCGM monitoring and supported telemetry exporters for centralized device diagnostics. But the present scenario is about a model endpoint experiences out-of-memory failures only during long prompts and concurrent sessions. For it, the needed control is to monitor used and available GPU memory across realistic serving contexts and concurrent workload sizes.