Cisco 300-640 Practice Test 17
Question 160
The acceptance board must decide whether an AI-infrastructure response to segmentation of model data traffic is fit for controlled operation. The fictional operating facts are: A cloud-edge cluster uses identical network labels for storage replication and contractor management traffic despite different trust levels. Signoff cannot rely on a good-looking utilization graph, a change ticket alone or an unmeasured promise. The response must include an appropriate governing engineering principle, a scoped intervention and a separate measurable acceptance criterion. Choose EXACTLY THREE statements that meet those roles.
- Coordinate facility-validated cooling distribution and thermal controls before increasing sustained GPU power.
- Training, management, tenant and sensitive data paths need enforceable policy boundaries independent of aggregate throughput.
- Authorized training jobs complete while isolation probes cannot reach disallowed resources.
- GPU duty cycle and samples-per-second improve with no host memory thrashing.
- Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access.
- Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth.
Explanation
Correct Answer
B:A cloud-edge cluster uses identical network labels for storage replication and contractor management traffic despite different trust levels. Training, management, tenant and sensitive data paths need enforceable policy boundaries independent of aggregate throughput. Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access.
C:A cloud-edge cluster uses identical network labels for storage replication and contractor management traffic despite different trust levels. Authorized training jobs complete while isolation probes cannot reach disallowed resources. Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access.
E:A cloud-edge cluster uses identical network labels for storage replication and contractor management traffic despite different trust levels. Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access. Training, management, tenant and sensitive data paths need enforceable policy boundaries independent of aggregate throughput.
Incorrect Answers
A:Coordinate facility-validated cooling distribution and thermal controls before increasing sustained GPU power. It does not implement the governing rule for this approval: A cloud-edge cluster uses identical network labels for storage replication and contractor management traffic despite different trust levels. Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access. Authorized training jobs complete while isolation probes cannot reach disallowed resources.
D:GPU duty cycle and samples-per-second improve with no host memory thrashing. It would not independently demonstrate that the control works: A cloud-edge cluster uses identical network labels for storage replication and contractor management traffic despite different trust levels. Authorized training jobs complete while isolation probes cannot reach disallowed resources. Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access.
F:Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth. It is not the rule that supports the proposed change and acceptance test: A cloud-edge cluster uses identical network labels for storage replication and contractor management traffic despite different trust levels. Training, management, tenant and sensitive data paths need enforceable policy boundaries independent of aggregate throughput. Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access.
Question 161
A data-center AI architecture committee considers gpu memory and model fit in its AI Infrastructure Components and Architecture review. At the affected site: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. The committee is deliberately not selecting a vendor product or applying a live change yet. It needs the foundational reasoning that makes the next design decision defensible. Which ONE engineering principle directly governs this situation? Choose ONE answer.
- Training, management, tenant and sensitive data paths need enforceable policy boundaries independent of aggregate throughput.
- Checkpoint systems must meet recovery-point and recovery-time needs with verified durable writes and restore validation.
- GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
- High-density GPU racks must fit measured branch-circuit, PDU and failover electrical budgets under realistic power transients.
Explanation
Correct Answer
C:A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
Incorrect Answers
A:Training, management, tenant and sensitive data paths need enforceable policy boundaries independent of aggregate throughput. It does not govern the condition described: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
B:Checkpoint systems must meet recovery-point and recovery-time needs with verified durable writes and restore validation. It does not govern the condition described: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
D:High-density GPU racks must fit measured branch-circuit, PDU and failover electrical budgets under realistic power transients. It does not govern the condition described: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
Question 162
An infrastructure diagnostician is investigating gpu memory and model fit for the AI Infrastructure Components and Architecture team. The reported incident reads: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Competing teams attribute the symptoms to different layers of the AI stack. Before accepting a hypothesis or touching production settings, the investigator must choose a discriminating set of measurements. Which ONE evidence package is MOST relevant? Choose ONE answer.
- Map VLAN or VRF boundaries, workload identities, firewall rules, east-west flows and attempted cross-tenant connections.
- Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
- Compare inlet temperatures, coolant flow where relevant, facility heat rejection, exhaust recirculation and throttling events.
- Review rack nameplate ratings, measured peak draw, dual-feed loading, breaker headroom and PSU redundancy.
Explanation
Correct Answer
B:A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
Incorrect Answers
A:Map VLAN or VRF boundaries, workload identities, firewall rules, east-west flows and attempted cross-tenant connections. That check would not establish the decisive facts for this condition: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
C:Compare inlet temperatures, coolant flow where relevant, facility heat rejection, exhaust recirculation and throttling events. That check would not establish the decisive facts for this condition: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
D:Review rack nameplate ratings, measured peak draw, dual-feed loading, breaker headroom and PSU redundancy. That check would not establish the decisive facts for this condition: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
Question 163
During a bounded AI-platform change-control review concerning gpu memory and model fit, the site owner reports: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. The proposed work must have a narrow implementation scope, first be validated against the failure evidence and avoid unrelated system-wide tuning. Which ONE intervention should the engineers pilot before any wider rollout? Choose ONE answer.
- Use versioned immutable datasets with controlled transfer promotion and reject stale job inputs.
- Compare like-for-like PUE baselines and optimize cooling or capacity with a documented energy measurement plan.
- Coordinate facility-validated cooling distribution and thermal controls before increasing sustained GPU power.
- Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
Explanation
Correct Answer
D:A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
Incorrect Answers
A:Use versioned immutable datasets with controlled transfer promotion and reject stale job inputs. That change would leave the reported problem unresolved: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
B:Compare like-for-like PUE baselines and optimize cooling or capacity with a documented energy measurement plan. That change would leave the reported problem unresolved: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
C:Coordinate facility-validated cooling distribution and thermal controls before increasing sustained GPU power. That change would leave the reported problem unresolved: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
Question 164
Two independent technical reviewers examine a AI Infrastructure Components and Architecture problem involving gpu memory and model fit. Their shared incident log states: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. One reviewer requires measurements capable of separating the observed problem from neighboring failure modes; the other requires a bounded action that follows those measurements. Choose EXACTLY TWO statements that together supply those two parts.
- Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
- Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
- Map node-to-rack, PDU, network-fabric and storage failure domains alongside worker-placement requirements.
- Inventory immutable datasets, intermediate artifacts, checkpoint rotations, compression and quota high-water marks.
- Package or shard files appropriately and scale the supported metadata path while preserving data semantics.
Explanation
Correct Answer
A:A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. Training or serving completes a representative workload with safe high-water memory headroom.
B:A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. Training or serving completes a representative workload with safe high-water memory headroom.
Incorrect Answers
C:Map node-to-rack, PDU, network-fabric and storage failure domains alongside worker-placement requirements. It would not provide the evidence needed for the paired diagnosis and follow-up: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. Training or serving completes a representative workload with safe high-water memory headroom.
D:Inventory immutable datasets, intermediate artifacts, checkpoint rotations, compression and quota high-water marks. It would not provide the evidence needed for the paired diagnosis and follow-up: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. Training or serving completes a representative workload with safe high-water memory headroom.
E:Package or shard files appropriately and scale the supported metadata path while preserving data semantics. It is not the scoped response supported by the accompanying evidence: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. Training or serving completes a representative workload with safe high-water memory headroom.
Question 165
The acceptance board must decide whether an AI-infrastructure response to gpu memory and model fit is fit for controlled operation. The fictional operating facts are: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Signoff cannot rely on a good-looking utilization graph, a change ticket alone or an unmeasured promise. The response must include an appropriate governing engineering principle, a scoped intervention and a separate measurable acceptance criterion. Choose EXACTLY THREE statements that meet those roles.
- Training or serving completes a representative workload with safe high-water memory headroom.
- High file-open rates can bottleneck metadata services even when aggregate data bandwidth is low.
- GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
- Authorized training jobs complete while isolation probes cannot reach disallowed resources.
- Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
- Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
Explanation
Correct Answer
A:A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Training or serving completes a representative workload with safe high-water memory headroom. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
C:A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
F:A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
Incorrect Answers
B:High file-open rates can bottleneck metadata services even when aggregate data bandwidth is low. It is not the rule that supports the proposed change and acceptance test: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
D:Authorized training jobs complete while isolation probes cannot reach disallowed resources. It would not independently demonstrate that the control works: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Training or serving completes a representative workload with safe high-water memory headroom. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
E:Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. It does not implement the governing rule for this approval: A research team budgets GPUs by peak TFLOPS yet a large transformer fails during optimizer-state allocation. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. Training or serving completes a representative workload with safe high-water memory headroom.
Question 166
A data-center AI architecture committee considers gpu memory and model fit in its AI Infrastructure Components and Architecture review. At the affected site: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. The committee is deliberately not selecting a vendor product or applying a live change yet. It needs the foundational reasoning that makes the next design decision defensible. Which ONE engineering principle directly governs this situation? Choose ONE answer.
- Checkpoint systems must meet recovery-point and recovery-time needs with verified durable writes and restore validation.
- GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
- Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.
- High-density GPU racks must fit measured branch-circuit, PDU and failover electrical budgets under realistic power transients.
Explanation
Correct Answer
B:An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
Incorrect Answers
A:Checkpoint systems must meet recovery-point and recovery-time needs with verified durable writes and restore validation. It does not govern the condition described: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
C:Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic. It does not govern the condition described: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
D:High-density GPU racks must fit measured branch-circuit, PDU and failover electrical budgets under realistic power transients. It does not govern the condition described: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
Question 167
An infrastructure diagnostician is investigating gpu memory and model fit for the AI Infrastructure Components and Architecture team. The reported incident reads: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Competing teams attribute the symptoms to different layers of the AI stack. Before accepting a hypothesis or touching production settings, the investigator must choose a discriminating set of measurements. Which ONE evidence package is MOST relevant? Choose ONE answer.
- Model concurrent requests, cold-start time, per-replica memory, autoscaling delay and one-node-failed capacity.
- Map VLAN or VRF boundaries, workload identities, firewall rules, east-west flows and attempted cross-tenant connections.
- Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
- Map node-to-rack, PDU, network-fabric and storage failure domains alongside worker-placement requirements.
Explanation
Correct Answer
C:An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
Incorrect Answers
A:Model concurrent requests, cold-start time, per-replica memory, autoscaling delay and one-node-failed capacity. That check would not establish the decisive facts for this condition: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
B:Map VLAN or VRF boundaries, workload identities, firewall rules, east-west flows and attempted cross-tenant connections. That check would not establish the decisive facts for this condition: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
D:Map node-to-rack, PDU, network-fabric and storage failure domains alongside worker-placement requirements. That check would not establish the decisive facts for this condition: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
Question 168
During a bounded AI-platform change-control review concerning gpu memory and model fit, the site owner reports: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. The proposed work must have a narrow implementation scope, first be validated against the failure evidence and avoid unrelated system-wide tuning. Which ONE intervention should the engineers pilot before any wider rollout? Choose ONE answer.
- Use an approved durable checkpoint repository with atomic or validated writes and restore tests.
- Benchmark representative dataset shards and tune supported striping, caching or storage tiers based on bottlenecks.
- Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay.
- Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
Explanation
Correct Answer
D:An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
Incorrect Answers
A:Use an approved durable checkpoint repository with atomic or validated writes and restore tests. That change would leave the reported problem unresolved: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
B:Benchmark representative dataset shards and tune supported striping, caching or storage tiers based on bottlenecks. That change would leave the reported problem unresolved: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
C:Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay. That change would leave the reported problem unresolved: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
Question 169
Two independent technical reviewers examine a AI Infrastructure Components and Architecture problem involving gpu memory and model fit. Their shared incident log states: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. One reviewer requires measurements capable of separating the observed problem from neighboring failure modes; the other requires a bounded action that follows those measurements. Choose EXACTLY TWO statements that together supply those two parts.
- Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
- Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
- Use workload-aware anti-affinity and recovery checkpoints across independent domains where supported.
- Check checkpoint frequency, write completion, replicas, integrity hashes and actual full-state restore time.
- Inventory immutable datasets, intermediate artifacts, checkpoint rotations, compression and quota high-water marks.
Explanation
Correct Answer
A:An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. Training or serving completes a representative workload with safe high-water memory headroom.
B:An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. Training or serving completes a representative workload with safe high-water memory headroom.
Incorrect Answers
C:Use workload-aware anti-affinity and recovery checkpoints across independent domains where supported. It is not the scoped response supported by the accompanying evidence: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. Training or serving completes a representative workload with safe high-water memory headroom.
D:Check checkpoint frequency, write completion, replicas, integrity hashes and actual full-state restore time. It would not provide the evidence needed for the paired diagnosis and follow-up: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. Training or serving completes a representative workload with safe high-water memory headroom.
E:Inventory immutable datasets, intermediate artifacts, checkpoint rotations, compression and quota high-water marks. It would not provide the evidence needed for the paired diagnosis and follow-up: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. Training or serving completes a representative workload with safe high-water memory headroom.
Question 170
The acceptance board must decide whether an AI-infrastructure response to gpu memory and model fit is fit for controlled operation. The fictional operating facts are: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Signoff cannot rely on a good-looking utilization graph, a change ticket alone or an unmeasured promise. The response must include an appropriate governing engineering principle, a scoped intervention and a separate measurable acceptance criterion. Choose EXACTLY THREE statements that meet those roles.
- Measured p99 inference response remains within budget during the approved peak-burst workload.
- Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
- Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth.
- Training or serving completes a representative workload with safe high-water memory headroom.
- Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
- GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
Explanation
Correct Answer
D:An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Training or serving completes a representative workload with safe high-water memory headroom. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
E:An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
F:An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
Incorrect Answers
A:Measured p99 inference response remains within budget during the approved peak-burst workload. It would not independently demonstrate that the control works: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Training or serving completes a representative workload with safe high-water memory headroom. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.
B:Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. It does not implement the governing rule for this approval: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use. Training or serving completes a representative workload with safe high-water memory headroom.
C:Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth. It is not the rule that supports the proposed change and acceptance test: An inference service selects a small-memory accelerator that loads the model but crashes at concurrent long-context requests. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. Revise supported precision, checkpointing or model sharding and select GPU memory capacity based on peak use.