Key details for this exam, checked against the published exam outline
Each question shows the correct answer and an explanation of why it is right
A data science team wants to deploy a customer support chatbot that answers questions using the organization's internal knowledge base documents without retraining the underlying large language model. The solution needs to retrieve relevant document snippets at query time and inject them into the model's context window before generating a response.
Which AI/ML workload type does this scenario describe?
The correct answer is RAG. RAG augments a generative model's responses by retrieving relevant external data (such as documents from a knowledge base) at query time and incorporating it into the prompt, without altering model weights. This differs from training, which involves adjusting model parameters over many epochs, and from fine-tuning, which updates weights on a smaller specialized dataset. Batch inference simply refers to running inference on data in bulk, not retrieval-based augmentation.
An enterprise wants to run latency-sensitive AI inference for a manufacturing quality-control camera system directly on the factory floor because sending high-resolution video to a centralized data center or public cloud would introduce unacceptable delay and consume excessive WAN bandwidth. Only lightweight inference is needed locally, while periodic model updates are pushed from a central location.
Which type of AI infrastructure deployment best fits this requirement?
Edge AI is correct because it places inference compute close to the data source (the factory floor cameras) to minimize latency and reduce bandwidth usage, while still allowing centralized model management. Public cloud and centralized on-premises deployments would introduce the latency and bandwidth issues described. A hybrid deployment with full workload mobility describes moving workloads between cloud and on-premises, which is not the primary requirement here—the key driver is proximity to the data source for low-latency inference.
A network architect is designing the fabric for a new GPU cluster that will run large-scale distributed training jobs using RDMA over Converged Ethernet. The design must minimize packet loss and ensure lossless transport across the fabric so that GPU-to-GPU communication does not stall during collective operations such as all-reduce.
Which combination of mechanisms should be implemented on the Cisco Nexus switches to achieve lossless Ethernet transport for this RoCEv2 traffic?
PFC, ECN, and ETS together form the standard congestion management toolkit for lossless Ethernet fabrics carrying RoCEv2 traffic. PFC pauses specific traffic classes to prevent buffer overflow, ECN marks packets to signal congestion before drops occur, and ETS allocates bandwidth guarantees across traffic classes. STP/VLAN pruning, uRPF/NAT, HSRP/VRRP, and BGP route reflection all serve unrelated purposes (loop prevention, security, redundancy, and routing scale) and do not provide lossless RDMA transport.
A Cisco UCS administrator is building a domain profile in Intersight for a new AI training cluster with GPU-equipped servers. The administrator needs the servers to receive accurate timestamps for distributed logging and telemetry correlation across the fabric, since GPU job schedulers rely on synchronized clocks to properly sequence training checkpoints.
Which Cisco UCS policy should be attached to the domain profile to address this requirement?
The NTP policy is used within a UCS domain profile in Intersight to configure Network Time Protocol servers so that all managed devices maintain synchronized clocks, which is essential for accurate log correlation and telemetry timestamps across distributed GPU training jobs. Power policy governs power capping and redundancy, QoS policy governs traffic classes and bandwidth allocation, and storage policy governs disk group and volume configuration—none of these address time synchronization.
An operations team monitoring an AI data center fabric with Cisco Nexus Dashboard notices repeated PFC pause frame counters incrementing on several interfaces connected to GPU servers, along with intermittent drops in RoCEv2 throughput reported by the training job. They want to correlate this network-level telemetry with switch buffer utilization and host-side NIC counters to isolate whether the bottleneck is caused by microbursts on a specific leaf switch uplink.
Which capability of Nexus Dashboard is most directly used to accomplish this cross-domain troubleshooting task?
Nexus Dashboard's operational telemetry and log correlation capabilities allow operators to aggregate and correlate data such as PFC counters, buffer utilization, and interface statistics across multiple devices to pinpoint the root cause of performance issues like microbursts. Zero-touch provisioning, VXLAN EVPN route reflector setup, and firmware upgrade scheduling are all valid Nexus Dashboard or fabric management functions, but none of them are used for correlating telemetry data to troubleshoot a live performance bottleneck.
60 questions covering all exam domains, starting from $20
Exam domains verified against: Official Cisco 300-640 exam guide, last checked September 2026.
Learn to describe AI and ML workload types including RAG, training, inference, and generative AI. Understand the AI lifecycle, use cases, infrastructure types from cloud to edge, and the network, compute, storage, orchestration and monitoring components that support AI environments.
Evaluate network, compute and storage deployments against AI workload requirements for bandwidth, latency, GPU resources, memory and scalability. Assess power efficiency and sustainability needs, and design hybrid deployments with secure cloud integration and data synchronization.
Configure high-performance networks using congestion control, RoCE and QoS to support AI workloads. Set up compute and storage with Cisco UCS domain profiles, power policies and connectivity policies, and deploy AI-ready fabrics using Nexus Dashboard, APIC and Intersight.
Implement benchmarks to measure AI infrastructure performance. Monitor data center health using Nexus Dashboard and Intersight to track operational telemetry, system health and alerts, then troubleshoot issues using system messages and log correlation tools.
Sample question from this domain above: Q5
Common questions about the exam itself
