300-640 Questions and Answers

Cisco's validated guidance for connecting VAST Data storage through an ACI fabric explicitly configures the RDMA storage class with Weighted Random Early Detection and Congestion Notification enabled . This makes option A directly consistent with Cisco's documented configuration.

In Cisco's VAST Data ACI design example, storage traffic arrives marked with DSCP 26 and is assigned to the Level2 QoS class. PFC is enabled for the corresponding no-drop DSCP, bandwidth is reserved for storage traffic, the congestion algorithm is configured as WRED, and congestion notification is enabled.

WRED and ECN-style congestion notification work together to detect developing queue pressure before buffers overflow. Rather than waiting for tail-drop conditions, the network can mark congestion-capable traffic so RoCEv2 endpoints reduce their transmission rates. PFC provides the complementary no-drop behavior when congestion becomes sufficiently severe.

Option B disables the required congestion signaling. Options C and D use Tail Drop, which Cisco's documented VAST RDMA storage class does not use. Tail Drop is instead shown in the same Cisco design for control-plane and ordinary traffic classes where lossless RDMA behavior is not required.

This configuration is therefore not merely a generic QoS preference; it directly follows Cisco's published ACI-to-VAST storage design.

Study Guide Reference: AI Infrastructure Deployment and Data Management — QoS, ECN/WRED, PFC, RDMA storage, and lossless fabric configuration.

Ethernet Network Control Policy

Ethernet Network Group Policy

Ethernet QoS Policy

Ethernet Adapter Policy

The mappings correspond to the functional boundaries of Cisco Intersight vNIC policies. Ethernet Network Control Policy manages link-layer behavior including CDP and LLDP. Cisco's Intersight documentation explicitly identifies CDP, LLDP transmit/receive, MAC registration, and uplink-failure behavior as Network Control settings.

Ethernet Network Group Policy controls VLAN membership for FI-attached servers, including the set of allowed VLANs and the optional native VLAN. Ethernet QoS Policy supplies performance-oriented vNIC characteristics such as MTU, rate limiting, traffic priority, CoS, and burst parameters. Cisco's current RoCE configuration guide lists MTU and rate-limit values directly under Ethernet QoS configuration.

Finally, Ethernet Adapter Policy contains adapter-level behavior and performance parameters, including RoCE configuration used for high-performance RDMA interfaces.

Ethernet Network Policy is the unused choice. Cisco documents that policy primarily for VLAN configuration on standalone servers, whereas Ethernet Network Group Policy is the relevant VLAN policy for Fabric Interconnect-attached UCS X-Series vNICs.

Study Guide Reference: AI Infrastructure Deployment and Data Management — UCS vNIC templates, Ethernet policies, QoS, VLAN configuration, and RoCE.

===============

The PFC pause-frame transmission rate per interface is the most directly relevant congestion indicator for an AI training fabric carrying RoCEv2 traffic. Priority Flow Control protects lossless traffic classes by transmitting pause frames when a receiving interface or switch queue approaches a congestion threshold. An elevated or sustained PFC pause rate therefore indicates that the lossless RDMA traffic class is repeatedly encountering queue pressure.

This is particularly significant for distributed AI training. GPU collective operations require synchronized communication between large numbers of accelerators, and excessive congestion can create tail latency, reduce effective network throughput, and extend Job Completion Time. Cisco AI networking guidance consequently treats PFC, ECN, queue occupancy, drops, and congestion telemetry as key operational signals for AI backend fabrics.

Spanning-tree topology changes are largely irrelevant in a modern routed or VXLAN-EVPN AI backend. BGP EVPN route counts are useful for control-plane capacity assessment but do not directly reveal congestion. VXLAN encapsulation rates measure traffic volume rather than whether buffers are becoming saturated.

PFC should not be interpreted as inherently problematic—some pause activity may occur during microbursts. However, unusually high or persistent pause-frame rates are a strong signal that congestion behavior requires investigation.

Study Guide Reference: AI Infrastructure Operations and Troubleshooting — RoCEv2 monitoring, PFC, congestion, telemetry, and AI training performance.

===============

Two settings determine the required behavior. First, Power Restore must be configured as Always On . Cisco defines this setting as instructing the server to power on after power is restored following a loss event. Immediate is not one of the documented Power Restore states; supported states include Last State, Always On, and Always Off.

Second, Power Priority should be High . Cisco explains that Power Priority determines initial power allocation among servers. A server assigned a higher priority receives preference when power is being allocated, which directly supports the requirement that these systems restart as soon as possible. Cisco validated designs explicitly show Power Profiling enabled with Power Priority set to High for performance-oriented deployments.

Power Profiling permits the platform to determine server power requirements. The Processor Package Power Limit should remain at Default for the X-Series scenario. Cisco documentation also notes that the selectable PPL setting is specifically applicable to particular C225/C245 M8 systems rather than being the mechanism for prioritizing X-Series restart behavior.

Option C provides Always On but fails to assign High power priority. Options A and D contain the invalid Immediate restore setting.

Study Guide Reference: AI Infrastructure Operations and Troubleshooting — UCS power management, recovery after power loss, power prioritization, and infrastructure availability.

In an Intersight Managed Mode UCS domain, DNS and NTP configuration is policy-controlled through the UCS Domain Profile deployed to the Fabric Interconnects. Changes entered locally through the Device Console can restore temporary reachability, but the policy configuration maintained by Intersight remains authoritative. If the Domain Profile still contains the previous DNS or NTP configuration, a subsequent deployment can overwrite the locally modified values.

Cisco documents that the Domain Profile references both the Network Connectivity Policy , which contains DNS settings, and the NTP Policy , which defines the NTP servers. The appropriate persistent remediation is therefore to edit the configuration associated with the deployed Domain Profile and redeploy it.

The IMC Access Policy applies primarily to server-management addressing and access rather than Fabric Interconnect DNS/NTP persistence. A Device Connector policy does not supply these domain DNS settings. Option C incorrectly applies a Network Connectivity Policy to the blade chassis; for the Fabric Interconnect domain, the relevant network-connectivity configuration is referenced through the UCS Domain Profile.

Cisco also advises that DNS and NTP should be configured on the management interface rather than being treated as Device Connector settings.

Study Guide Reference: AI Infrastructure Deployment and Data Management — Intersight Managed Mode, UCS Domain Profiles, DNS/NTP policies, and management connectivity.

Rapid provisioning supports scalability by reducing the time required to make compute, GPU, networking, storage, and orchestration resources available for workloads. The most appropriate description is therefore the efficient launch of computational platforms . When new AI capacity can be provisioned consistently and quickly, organizations can expand training environments, increase inference capacity, respond to workload bursts, and introduce additional accelerator resources without lengthy manual configuration.

Cisco's DCAI objectives explicitly cover orchestration as a component of AI environments and require candidates to evaluate infrastructure scalability. Cisco also emphasizes that AI infrastructure must scale to accommodate changing workloads and that cloud-oriented infrastructure offers rapid access to computational resources when demand changes.

Option A introduces fragmented management, which generally increases operational complexity. Option B restricts connectivity and therefore conflicts with scalable distributed AI execution. Option C explicitly delays resource availability and is the opposite of rapid provisioning.

The architectural objective is to make infrastructure consumption repeatable and automated. Efficient provisioning allows the underlying platform to expand at the pace demanded by GPU-intensive training and production inference rather than becoming an operational bottleneck.

Study Guide Reference: 1.0 AI Fundamentals and Applications — 1.5.d Orchestration; AI infrastructure scalability concepts.

===============

backend network

frontend network

storage network

management network

Large AI environments normally separate traffic into purpose-built fabrics because GPU collective communication, storage I/O, application access, and infrastructure management have fundamentally different performance and availability characteristics. Cisco identifies the backend network as the east-west fabric dedicated to inter-GPU communication. It carries high-bandwidth RoCEv2/RDMA traffic associated with distributed training and must provide exceptionally low latency and lossless behavior.

The frontend network provides north-south connectivity between AI nodes and the wider data center, supporting application access, inference traffic, general data movement, logging, and other in-band services. The storage network connects compute nodes to shared datasets, model checkpoints, and high-performance parallel or distributed storage systems. Cisco AI POD designs may use dedicated storage-leaf connectivity to prevent storage traffic from interfering with backend GPU collective communication. The management network serves out-of-band management interfaces for servers, switches, storage platforms, and other infrastructure components. Cisco AI reference architectures explicitly distinguish backend, frontend/storage, and management connectivity for these purposes.

Maintaining these traffic domains independently improves deterministic performance, fault isolation, congestion control, security, and operational manageability.

Study Guide Reference: AI Infrastructure Components and Architecture — network fabrics, bandwidth, latency, scalability, and AI infrastructure connectivity.

===============

The Cisco UCS C885A M8 with NVIDIA H200 GPUs is purpose-built for dense accelerated computing and directly satisfies the NVLink requirement. Cisco's C885A M8 NVIDIA configuration supports eight H100 or H200 SXM Tensor Core GPUs on an HGX platform. Those GPUs are interconnected using NVIDIA NVLink, providing the high-bandwidth GPU-to-GPU communication necessary for distributed tensor processing inside the server.

The platform also provides substantial host-memory capacity, dual AMD EPYC processors, high-speed local NVMe storage, multiple 400-GbE adapters for inter-node GPU communication, and Cisco Intersight integration. These capabilities make it appropriate for large language models, generative AI training, fine-tuning, RAG, and large inference workloads.

Option A does not provide the same HGX NVLink-based eight-GPU architecture with an L40 configuration. Option B uses AMD accelerators; AMD GPU connectivity employs AMD interconnect technologies rather than NVIDIA NVLink. Option C is designed for materially different workload and GPU-density requirements and does not deliver the HGX H200 NVLink topology specified by the scenario.

The critical selection criterion is not merely whether a server can host a discrete GPU. The question explicitly requires direct GPU-to-GPU access through NVLink , which identifies the NVIDIA HGX-based C885A M8 configuration.

Study Guide Reference: AI Infrastructure Components and Architecture — compute selection based on CPU, GPU, memory, virtualization, interconnect performance, and workload requirements.

===============

Traffic Analytics and Flow Telemetry are mutually exclusive flow-collection modes within Cisco Nexus Dashboard. Cisco's current Nexus Dashboard documentation identifies both under Flow Collection settings and states that Flow Telemetry must be disabled at the fabric level before Traffic Analytics can be enabled.

Traffic Analytics automatically discovers services and visualizes conversations based primarily on Layer 4 information. It provides visibility into network characteristics such as congestion, latency, and drops, which directly matches the monitoring requirements in the scenario. Cisco's Nexus Dashboard Insights guidance likewise states that if Flow Telemetry is already active, it must first be disabled for the relevant sites before Traffic Analytics is selected.

Delta Analysis is an assurance/comparison capability and is not the mutually exclusive collection mode described here. Connectivity Analysis performs reachability and path-oriented diagnostics. SNMP traps remain a separate event-notification mechanism and are not displaced by Traffic Analytics.

The operational distinction matters because Traffic Analytics and Flow Telemetry use different approaches to flow collection and analysis. An administrator cannot simply enable both collection modes simultaneously and expect them to operate in parallel on the same fabric configuration.

Study Guide Reference: AI Infrastructure Operations and Troubleshooting — Nexus Dashboard monitoring, telemetry, congestion analysis, latency, and packet-drop visibility.

===============

NVIDIA Collective Communications Library (NCCL) is the appropriate framework for validating communication performance between NVIDIA GPU nodes. NCCL provides highly optimized implementations of collective operations such as AllReduce, AllGather, Reduce, Broadcast, and ReduceScatter. These operations closely represent the communication patterns generated by distributed AI training and therefore provide much more meaningful validation of GPU-cluster performance than conventional network-throughput tests.

NCCL-based benchmarks can reveal whether GPU-to-GPU communication is achieving expected bandwidth and latency across NVLink, PCIe, and the external backend network. They are particularly useful when validating RoCEv2 fabrics because distributed training performance depends on collective communication efficiency rather than simple host-to-host TCP throughput.

iPerf measures conventional TCP or UDP network performance between hosts and is useful for basic network validation, but it does not exercise NVIDIA GPU collective communications. Spirent and Ixia provide sophisticated packet-generation and network-testing platforms, yet they test the network primarily from a packet-forwarding perspective rather than measuring actual GPU collective-operation behavior.

For a newly deployed NVIDIA AI cluster, NCCL benchmarking therefore provides workload-representative validation of the end-to-end GPU communication architecture and can expose problems involving topology, NIC configuration, congestion control, bandwidth imbalance, or fabric design.

Study Guide Reference: AI Infrastructure Operations and Troubleshooting — performance benchmarking, NVIDIA GPU communications, backend-fabric validation, and distributed AI workload testing.

===============

Cisco Nexus 9000 integration with NVIDIA adaptive-routing capabilities requires the Nexus switch to enable the Spectrum-X hardware profile . Cisco's current NX-OS configuration guide explicitly specifies the global configuration command hardware profile spectrum-x to enable NVIDIA Spectrum-X adaptive routing. When a Nexus switch peers with a supported NVIDIA NIC, this capability allows the NIC to handle packet reordering created by granular techniques such as per-packet Dynamic Load Balancing.

This is particularly important for AI training because large synchronized RDMA flows can create uneven utilization under conventional hash-based ECMP. Cisco explains that per-packet DLB can spray packets over multiple paths and that NVIDIA adaptive routing/direct data placement enables the receiver to correctly handle packets that arrive out of order, improving utilization and collective-operation performance.

Option A is not the documented NX-OS configuration syntax. Option B is not the Cisco-prescribed Nexus integration command. Option D may be relevant to traffic classification in other designs but does not activate Spectrum-X adaptive-routing interoperability.

Therefore, the Cisco-documented configuration is C. This differs from the supplied file's marked answer B.

Study Guide Reference: 3.0 AI Infrastructure Deployment and Data Management — 3.1.d Load distribution.

===============

Congestion Notification Packets are control-feedback messages used by RoCEv2 congestion-management mechanisms. When a receiving endpoint observes an ECN-marked packet, it generates a CNP toward the sender, instructing the transmitting endpoint to reduce its transmission rate. Losing these packets would delay congestion response and allow queues to continue building.

Cisco's validated RoCEv2-over-VXLAN design explicitly states that CNP packets use DSCP 48 and are classified into a high-priority queue . In Cisco's example, ordinary RoCEv2 traffic is mapped to QoS group 3, while CNP is mapped to QoS group 7. Cisco further states that the high-priority classification ensures that CNP packets are not dropped in the network.

Traffic shaping in option A would regulate CNP transmission rather than guarantee timely feedback. Option C places CNP in the same potentially congested queue as the data traffic that generated the congestion, defeating the purpose of reliable feedback. Option D is also incorrect because PFC is intended to protect a lossless data priority; CNP itself should be forwarded through a high-priority queue and should not depend on pause behavior.

Study Guide Reference: AI Infrastructure Deployment and Data Management — RoCEv2, ECN, CNP, QoS classification, and congestion management.

===============

The intended solution is Cisco Nexus Hyperfabric AI because it provides an integrated on-premises AI infrastructure managed through a cloud-hosted controller. Cisco's current reference architecture documents a configuration using 12 Cisco UCS C885A M8 servers with eight NVIDIA GPUs each, producing a 96-GPU cluster . The architecture uses Cisco Hyperfabric switching and is engineered for high-performance AI training and inference.

Cisco also states that the full-stack Hyperfabric AI infrastructure is aligned with NVIDIA Enterprise Reference Architecture requirements and uses a cloud controller to manage network design, deployment, monitoring, upgrades, troubleshooting, and scaling.

This eliminates option C because Nexus Dashboard is not the cloud-managed Hyperfabric full-stack architecture described. Option D incorrectly characterizes Intersight as the AI network-fabric controller; Intersight manages UCS infrastructure, whereas Nexus Hyperfabric provides the required cloud-managed network lifecycle.

The uploaded question specifically maps the 96-GPU requirement to the Small AI Cluster , and its source answer is A. Current Cisco public documentation confirms the 96-GPU Hyperfabric topology and cloud-managed NVIDIA-aligned architecture, although current product material presents scalable cluster templates more generally as small through extra-large.

Study Guide Reference: 3.3.c Hyperfabric; 2.1 network scalability; 2.2 GPU compute scalability.