cert
cert-1
cert-2

Pass Cisco 300-640 Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers to Pass the Actual Exam!
30 Days Free Updates, Instant Download!

cert-5
cert-6
300-640 Exam - Verified By Experts
300-640 Premium File

300-640 Premium File

$89.99
$98.99
  • Premium File 60 Questions & Answers. Last Update: Nov 10, 2026

Whats Included:

  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates
 
$98.99
$89.99
accept 10 downloads in the last 7 days
block-screenshots
Cisco 300-640 Practice Test 13

Question 120

The acceptance board must decide whether an AI-infrastructure response to validated cisco ai infrastructure patterns is fit for controlled operation. The fictional operating facts are: A healthcare provider chooses a managed AI fabric workflow but assumes every switch firmware combination is automatically supported by the selected Cisco orchestration release. 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.

  1. The as-built configuration satisfies the approved support matrix and workload acceptance benchmarks.
  2. Compare the planned cluster with the current validated Cisco reference architecture and test deviations in a controlled pilot.
  3. Edge AI should tolerate intermittent upstream connectivity while honoring local compute, power, update and privacy limits.
  4. A vendor reference design is a starting constraint set; validate the exact Cisco AI POD or fabric bill of materials, software support and workload fit before implementation.
  5. Pilot the narrowly scoped inference workflow with a named owner and benchmark it against a simple deterministic baseline.
  6. A staged test proves authorized transfer, version parity and the documented loss-of-connectivity behavior.

Explanation

Correct Answer

A:

A healthcare provider chooses a managed AI fabric workflow but assumes every switch firmware combination is automatically supported by the selected Cisco orchestration release. The as-built configuration satisfies the approved support matrix and workload acceptance benchmarks. Compare the planned cluster with the current validated Cisco reference architecture and test deviations in a controlled pilot.

B:

A healthcare provider chooses a managed AI fabric workflow but assumes every switch firmware combination is automatically supported by the selected Cisco orchestration release. Compare the planned cluster with the current validated Cisco reference architecture and test deviations in a controlled pilot. A vendor reference design is a starting constraint set; validate the exact Cisco AI POD or fabric bill of materials, software support and workload fit before implementation.

D:

A healthcare provider chooses a managed AI fabric workflow but assumes every switch firmware combination is automatically supported by the selected Cisco orchestration release. A vendor reference design is a starting constraint set; validate the exact Cisco AI POD or fabric bill of materials, software support and workload fit before implementation. Compare the planned cluster with the current validated Cisco reference architecture and test deviations in a controlled pilot.

Incorrect Answers

C:

Edge AI should tolerate intermittent upstream connectivity while honoring local compute, power, update and privacy limits. It is not the rule that supports the proposed change and acceptance test: A healthcare provider chooses a managed AI fabric workflow but assumes every switch firmware combination is automatically supported by the selected Cisco orchestration release. A vendor reference design is a starting constraint set; validate the exact Cisco AI POD or fabric bill of materials, software support and workload fit before implementation. Compare the planned cluster with the current validated Cisco reference architecture and test deviations in a controlled pilot.

E:

Pilot the narrowly scoped inference workflow with a named owner and benchmark it against a simple deterministic baseline. It does not implement the governing rule for this approval: A healthcare provider chooses a managed AI fabric workflow but assumes every switch firmware combination is automatically supported by the selected Cisco orchestration release. Compare the planned cluster with the current validated Cisco reference architecture and test deviations in a controlled pilot. The as-built configuration satisfies the approved support matrix and workload acceptance benchmarks.

F:

A staged test proves authorized transfer, version parity and the documented loss-of-connectivity behavior. It would not independently demonstrate that the control works: A healthcare provider chooses a managed AI fabric workflow but assumes every switch firmware combination is automatically supported by the selected Cisco orchestration release. The as-built configuration satisfies the approved support matrix and workload acceptance benchmarks. Compare the planned cluster with the current validated Cisco reference architecture and test deviations in a controlled pilot.

Question 121

A data-center AI architecture committee considers east-west training bandwidth in its AI Infrastructure Components and Architecture review. At the affected site: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. 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.

  1. Storage capacity plans require dataset lineage, scratch amplification, checkpoint retention and growth forecasts, not initial dataset size only.
  2. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.
  3. Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth.
  4. GPU partitioning and virtualization require platform-licensed, driver-compatible capabilities verified for the target workload.

Explanation

Correct Answer

B:

A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

Incorrect Answers

A:

Storage capacity plans require dataset lineage, scratch amplification, checkpoint retention and growth forecasts, not initial dataset size only. It does not govern the condition described: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

C:

Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth. It does not govern the condition described: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

D:

GPU partitioning and virtualization require platform-licensed, driver-compatible capabilities verified for the target workload. It does not govern the condition described: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

Question 122

An infrastructure diagnostician is investigating east-west training bandwidth for the AI Infrastructure Components and Architecture team. The reported incident reads: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. 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.

  1. Compare dataset manifests, content hashes, model versions, replication lag and divergent write histories.
  2. Check checkpoint frequency, write completion, replicas, integrity hashes and actual full-state restore time.
  3. Count metadata operations, open latency, namespace contention and worker-level file listing times.
  4. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

Explanation

Correct Answer

D:

A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

Incorrect Answers

A:

Compare dataset manifests, content hashes, model versions, replication lag and divergent write histories. That check would not establish the decisive facts for this condition: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

B:

Check checkpoint frequency, write completion, replicas, integrity hashes and actual full-state restore time. That check would not establish the decisive facts for this condition: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

C:

Count metadata operations, open latency, namespace contention and worker-level file listing times. That check would not establish the decisive facts for this condition: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

Question 123

During a bounded AI-platform change-control review concerning east-west training bandwidth, the site owner reports: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. 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.

  1. Compare like-for-like PUE baselines and optimize cooling or capacity with a documented energy measurement plan.
  2. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
  3. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.
  4. Reserve and prewarm sufficient inference replicas across independent hosts to meet stated SLO during failover.

Explanation

Correct Answer

C:

A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

Incorrect Answers

A:

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 pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

B:

Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. That change would leave the reported problem unresolved: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

D:

Reserve and prewarm sufficient inference replicas across independent hosts to meet stated SLO during failover. That change would leave the reported problem unresolved: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

Question 124

Two independent technical reviewers examine a AI Infrastructure Components and Architecture problem involving east-west training bandwidth. Their shared incident log states: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. 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.

  1. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.
  2. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.
  3. Compare dataset manifests, content hashes, model versions, replication lag and divergent write histories.
  4. Coordinate facility-validated cooling distribution and thermal controls before increasing sustained GPU power.
  5. Compare inlet temperatures, coolant flow where relevant, facility heat rejection, exhaust recirculation and throttling events.

Explanation

Correct Answer

A:

A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

B:

A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

Incorrect Answers

C:

Compare dataset manifests, content hashes, model versions, replication lag and divergent write histories. It would not provide the evidence needed for the paired diagnosis and follow-up: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

D:

Coordinate facility-validated cooling distribution and thermal controls before increasing sustained GPU power. It is not the scoped response supported by the accompanying evidence: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

E:

Compare inlet temperatures, coolant flow where relevant, facility heat rejection, exhaust recirculation and throttling events. It would not provide the evidence needed for the paired diagnosis and follow-up: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

Question 125

The acceptance board must decide whether an AI-infrastructure response to east-west training bandwidth is fit for controlled operation. The fictional operating facts are: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. 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.

  1. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
  2. GPU duty cycle and samples-per-second improve with no host memory thrashing.
  3. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.
  4. Use versioned immutable datasets with controlled transfer promotion and reject stale job inputs.
  5. The multi-node training benchmark meets required epoch time without sustained fabric congestion.
  6. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

Explanation

Correct Answer

C:

A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

E:

A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. The multi-node training benchmark meets required epoch time without sustained fabric congestion. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

F:

A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

Incorrect Answers

A:

GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. It is not the rule that supports the proposed change and acceptance test: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

B:

GPU duty cycle and samples-per-second improve with no host memory thrashing. It would not independently demonstrate that the control works: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. The multi-node training benchmark meets required epoch time without sustained fabric congestion. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

D:

Use versioned immutable datasets with controlled transfer promotion and reject stale job inputs. It does not implement the governing rule for this approval: A pharmaceutical lab expands from eight to thirty-two GPU workers but keeps a shared narrow leaf uplink, making gradient exchanges dominate epochs. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

Question 126

A data-center AI architecture committee considers east-west training bandwidth in its AI Infrastructure Components and Architecture review. At the affected site: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. 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.

  1. Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth.
  2. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.
  3. Efficiency proposals must distinguish total facility power from IT energy and compare measured PUE under equivalent loads.
  4. Checkpoint systems must meet recovery-point and recovery-time needs with verified durable writes and restore validation.

Explanation

Correct Answer

B:

An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

Incorrect Answers

A:

Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth. It does not govern the condition described: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

C:

Efficiency proposals must distinguish total facility power from IT energy and compare measured PUE under equivalent loads. It does not govern the condition described: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

D:

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 ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

Question 127

An infrastructure diagnostician is investigating east-west training bandwidth for the AI Infrastructure Components and Architecture team. The reported incident reads: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. 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.

  1. Collect calibrated facility and IT power readings, interval timestamps, seasonal conditions and utilization context.
  2. Compare dataset manifests, content hashes, model versions, replication lag and divergent write histories.
  3. Check firmware, drivers, hypervisor or container runtime compatibility, partition inventory and isolation tests.
  4. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

Explanation

Correct Answer

D:

An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

Incorrect Answers

A:

Collect calibrated facility and IT power readings, interval timestamps, seasonal conditions and utilization context. That check would not establish the decisive facts for this condition: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

B:

Compare dataset manifests, content hashes, model versions, replication lag and divergent write histories. That check would not establish the decisive facts for this condition: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

C:

Check firmware, drivers, hypervisor or container runtime compatibility, partition inventory and isolation tests. That check would not establish the decisive facts for this condition: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

Question 128

During a bounded AI-platform change-control review concerning east-west training bandwidth, the site owner reports: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. 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.

  1. Package or shard files appropriately and scale the supported metadata path while preserving data semantics.
  2. Right-size and distribute accelerator nodes across validated A/B feeds with capped power policy where required.
  3. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.
  4. Tune CPU worker allocation, memory pinning and NUMA or device placement before buying additional GPUs.

Explanation

Correct Answer

C:

An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.

Incorrect Answers

A:

Package or shard files appropriately and scale the supported metadata path while preserving data semantics. That change would leave the reported problem unresolved: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

B:

Right-size and distribute accelerator nodes across validated A/B feeds with capped power policy where required. That change would leave the reported problem unresolved: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

D:

Tune CPU worker allocation, memory pinning and NUMA or device placement before buying additional GPUs. That change would leave the reported problem unresolved: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

Question 129

Two independent technical reviewers examine a AI Infrastructure Components and Architecture problem involving east-west training bandwidth. Their shared incident log states: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. 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.

  1. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs.
  2. Review rack nameplate ratings, measured peak draw, dual-feed loading, breaker headroom and PSU redundancy.
  3. Compare dataset manifests, content hashes, model versions, replication lag and divergent write histories.
  4. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.
  5. Reserve and prewarm sufficient inference replicas across independent hosts to meet stated SLO during failover.

Explanation

Correct Answer

A:

An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

D:

An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

Incorrect Answers

B:

Review rack nameplate ratings, measured peak draw, dual-feed loading, breaker headroom and PSU redundancy. It would not provide the evidence needed for the paired diagnosis and follow-up: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

C:

Compare dataset manifests, content hashes, model versions, replication lag and divergent write histories. It would not provide the evidence needed for the paired diagnosis and follow-up: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Compare per-link utilization, incast windows, bisection demand, NIC speed and all-reduce job profiles during representative epochs. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

E:

Reserve and prewarm sufficient inference replicas across independent hosts to meet stated SLO during failover. It is not the scoped response supported by the accompanying evidence: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

Question 130

The acceptance board must decide whether an AI-infrastructure response to east-west training bandwidth is fit for controlled operation. The fictional operating facts are: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. 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.

  1. The multi-node training benchmark meets required epoch time without sustained fabric congestion.
  2. GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
  3. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.
  4. Multiple tenants operate within their allocations while a full-device benchmark remains unaffected.
  5. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.
  6. Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access.

Explanation

Correct Answer

A:

An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. The multi-node training benchmark meets required epoch time without sustained fabric congestion. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

C:

An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

E:

An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic.

Incorrect Answers

B:

GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. It is not the rule that supports the proposed change and acceptance test: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Distributed training requires adequate effective east-west throughput between GPU workers, with oversubscription based on real collective traffic. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

D:

Multiple tenants operate within their allocations while a full-device benchmark remains unaffected. It would not independently demonstrate that the control works: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. The multi-node training benchmark meets required epoch time without sustained fabric congestion. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes.

F:

Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access. It does not implement the governing rule for this approval: An ad-ranking training farm shows strong single-node step times but cross-rack all-reduce performance collapses during job overlap. Size and balance the leaf-spine fabric for the observed communication pattern before adding accelerator nodes. The multi-node training benchmark meets required epoch time without sustained fabric congestion.

Implementing Cisco Data Center AI Infrastructure (DCAI) (300-640) Practice Tests