Latest Jul-2026 Cisco 300-640 Dumps Updated 82 Questions
PDF Download Free of 300-640 Valid Practice Test Questions
NEW QUESTION # 34
A customer is deploying a new AI fabric with all new NVIDIA GPUs and must verify the performance between the nodes in the network. Which tool must be used to validate the customer benchmarks?
- A. iPerf
- B. NCCL
- C. IXIA
- D. Spirent
Answer: B
Explanation:
NCCL is the standard tool used to validate GPU-to-GPU communication performance in NVIDIA- based AI fabrics. It measures collective communication behavior between GPU nodes, making it appropriate for benchmarking the AI training network performance.
NEW QUESTION # 35
Which result is provided through image recognition using AI?
- A. detection and labeling of visual elements from digital sources
- B. routing of multicast messages across network boundaries
- C. anonymization of audit trails for compliance objectives
- D. scheduling of management tasks during maintenance periods
Answer: A
Explanation:
Image recognition uses AI to analyze digital images or video frames and identify, detect, classify, or label visual elements such as objects, people, patterns, or scenes.
NEW QUESTION # 36
What is a function of NVLink technology within a high-performance AI server or cluster?
- A. to provide a high-speed, low-latency direct interconnect for GPU-to-GPU communication
- B. to enable remote access and virtualization of GPU resources for distributed AI development teams
- C. to establish a secure, encrypted channel for data transfer between the AI server and external storage
- D. to manage power distribution and thermal regulation across all installed GPU units
Answer: A
Explanation:
NVLink provides a high-bandwidth, low-latency direct interconnect between GPUs. In AI servers and clusters, it accelerates GPU-to-GPU communication for training and inference workloads by allowing GPUs to exchange data much faster than through standard PCIe paths.
NEW QUESTION # 37
Which set of statements describes Quantized Congestion Notification?
- A. Priority Flow Control reacts first to mitigate congestion, and Explicit Congestion Notification acts as a fail-safe to prevent traffic drops if Priority Flow Control is insufficient.
A congestion notification packet sent toward the source by hosts and the network switch in response to receiving Explicit Congestion Notification with congestion experienced bits set. - B. Explicit Congestion Notification reacts first to mitigate congestion, and Priority Flow Control acts as a fail-safe to prevent traffic drops if Explicit Congestion Notification is insufficient.
A congestion notification packet sent toward the source by hosts and the network switch in response to receiving Explicit Congestion Notification with congestion experienced bits set. - C. Explicit Congestion Notification reacts first to mitigate congestion, and Priority Flow Control acts as a fail-safe to prevent traffic drops if Explicit Congestion Notification is insufficient.
A congestion notification packet sent toward the source by a host in response to receiving Explicit Congestion Notification with congestion experienced bits set. - D. Priority Flow Control reacts first to mitigate congestion, and Explicit Congestion Notification acts as a fail-safe to prevent traffic drops if Priority Flow Control is insufficient.
A congestion notification packet sent toward the source by a network switch in response to receiving Explicit Congestion Notification with congestion experienced bits set.
Answer: C
Explanation:
Quantized Congestion Notification relies on Explicit Congestion Notification as the primary congestion signal so the sender can reduce its transmission rate before packet loss occurs.
Priority Flow Control acts as a safety mechanism if congestion becomes severe, and the congestion notification packet is generated by the receiving host when it receives packets marked with congestion-experienced bits.
NEW QUESTION # 38
What is a primary behavior that occurs during the model training stage of the AI lifecycle?
- A. adjusting internal weights and biases
- B. embedding the model into systems with APIs
- C. establishing success criteria
- D. detecting model drift
Answer: A
Explanation:
During model training, the model learns from training data by iteratively adjusting its internal weights and biases to reduce prediction error and improve performance against the defined objective function.
NEW QUESTION # 39
An end customer plans to deploy an AI fabric in a new data center. Their engineers are more familiar with CLI-based networks and require the capability to automate standardized configuration templates and monitor flows for the fabric. Which data center automation tool meets the requirements?
- A. Digital Network Architecture Center
- B. HyperFabric
- C. Intersight
- D. Nexus Dashboard
Answer: D
Explanation:
Nexus Dashboard is the Cisco data center automation and operations platform used with Nexus fabrics. It supports standardized fabric configuration through automation templates and provides visibility features such as flow monitoring, making it the appropriate tool for engineers who need CLI-oriented data center fabric operations with automation and telemetry.
NEW QUESTION # 40
A Cisco AI infrastructure requires fast convergence after a spine switch failure. Which routing design principle best supports this objective?
- A. Single-path forwarding
- B. Layer 2 loop dependency
- C. Equal-cost multipath routing
- D. Static route redistribution
Answer: C
Explanation:
ECMP provides multiple active forwarding paths and enables rapid failover when a path becomes unavailable. AI environments benefit from resilient high-bandwidth forwarding. Static redistribution slows operational flexibility, while single-path forwarding and Layer 2 dependency reduce scalability and convergence efficiency.
NEW QUESTION # 41
Refer to the exhibit. Users report delayed response times and network latency when running certain AI workloads on a specific Cisco UCS Intersight Managed Mode Domain. Based on the fault message and reported symptoms, which troubleshooting action must be taken first to address the network latency issues?
- A. Review the AI workload configuration to optimize resource allocation (CPU, memory, GPU) for the affected applications.
- B. Review the pause frame counters for signs of congestion on Fabric Interconnect B Port 11.
- C. Examine the UCS fabric interconnect configuration for port-channel 11 on Fabric Interconnect B.
- D. Check the physical cabling and SFP+ transceivers on port 11 of switch B for potential hardware failures or incorrect configurations.
Answer: D
Explanation:
The fault shows a receive error statistics threshold crossing on switch B port 11. Because receive errors on a physical interface commonly indicate cabling, transceiver, or physical-layer issues, the first troubleshooting step is to inspect the physical connection and SFP+ components on the affected port before investigating workload or higher-layer configuration.
NEW QUESTION # 42
A network architect wants to provide high availability for AI compute nodes connected to dual Cisco Nexus switches. Which technology enables active-active uplink forwarding without STP blocking?
- A. VRRP
- B. GLBP
- C. PVST+
- D. vPC
Answer: D
Explanation:
Virtual Port Channel (vPC) allows devices to use active-active uplinks across two Nexus switches while avoiding spanning-tree blocking. This increases bandwidth utilization and resiliency for AI clusters. GLBP and VRRP provide gateway redundancy, while PVST+ still relies on blocking redundant paths.
NEW QUESTION # 43
A medium-sized breadmaking operation must deploy a fleet of GPU servers that meet these requirements for their flagship application:
- policy managed through cloud-based interface
- low latency between specific pairs of GPUs
- at least 2 GPUs per server
- all GPUs should be same model
- Intel CPU
- well suited for Inferencing
- customer is power conscious
Which server meets the requirements?
- A. C885A M8
- B. C245 M8
- C. C845A M8
- D. X210C M8 with x580 PCIe node
Answer: D
Explanation:
The UCS X210c M8 with the X580 PCIe node fits the requirements because it supports cloud- based policy management through Cisco Intersight, uses Intel Xeon processors, supports multiple same-model GPUs, and is well suited for GPU-accelerated inference workloads while allowing modular scale-out as demand grows. Cisco describes the X580p PCIe node as supporting up to four GPUs with X210c/X215c compute nodes for workloads including inference.
Reference:
https://www.cisco.com/c/en/us/products/collateral/servers-unified-computing/ucs-x-series- modular-system/mlperf-inference-ucs-x580p-wp.html
https://www.cisco.com/c/dam/en/us/products/collateral/servers-unified-computing/ucs-x-series- modular-system/x210cm8-specsheet.pdf
NEW QUESTION # 44
A network operations team deploys HyperFabric AI and wants to monitor fabric health to proactively identify performance issues. The team must track metrics specifically relevant to AI training workloads. Which metric is most critical for identifying congestion issues that could impact AI training performance in a HyperFabric AI deployment?
- A. spanning tree topology change notifications
- B. BGP EVPN route count per leaf switch
- C. VXLAN tunnel packet encapsulation rate
- D. PFC pause frame transmission rate per interface
Answer: D
Explanation:
PFC pause frame transmission rate is the key congestion indicator for lossless Ethernet fabrics used by AI training workloads. A high or increasing pause frame rate shows that interfaces are experiencing congestion and applying flow control, which can directly affect GPU-to-GPU communication performance in a HyperFabric AI environment.
NEW QUESTION # 45
An engineer must monitor congestion, flow latency, and traffic drops using Cisco Nexus Dashboard and chooses to use the Traffic Analytics feature. Which Nexus Dashboard feature is automatically disabled when Traffic Analytics is enabled?
- A. SNMP Traps
- B. Delta Analysis
- C. Connectivity Analysis
- D. Flow Telemetry
Answer: D
Explanation:
Traffic Analytics and Flow Telemetry use overlapping telemetry resources in Cisco Nexus Dashboard. When Traffic Analytics is enabled to monitor congestion, latency, and drops, Flow Telemetry is automatically disabled to avoid resource and data collection conflicts.
NEW QUESTION # 46
A network engineer analyzes RoCEv2 traffic flows in an AI training cluster and notices that all RoCEv2 packets have the Don't Fragment bit set in the IP header. Why must RoCEv2 set the DF bit in all packets?
- A. to prevent IP fragmentation that would break the InfiniBand sequence numbering and ordering guarantees
- B. to enable Path MTU Discovery so that the optimal packet size can be negotiated
- C. to ensure that packets always take the same path through ECMP routing for consistent latency
- D. to indicate to intermediate switches that the packet contains RDMA traffic requiring special handling
Answer: A
Explanation:
RoCEv2 carries InfiniBand transport semantics over UDP/IP, and IP fragmentation would disrupt the expected packet sequencing and ordering behavior required by RDMA. Setting the Don't Fragment bit prevents intermediate devices from fragmenting RoCEv2 packets, preserving reliable RDMA transport behavior across the AI training fabric.
NEW QUESTION # 47
An organization deploys a new AI training fabric that uses RoCEv2 for GPU communication. The network architect designs the QoS configuration to ensure reliable RDMA transport and must meet these requirements:
- Support 256 GPU servers with RDMA connectivity.
- Prevent any packet loss that causes RDMA connection failures.
- Maintain consistent low-latency communication with a target of less
than 10 microseconds.
- Use industry-standard protocols and configurations.
Which configuration ensures that RoCEv2 operates as a lossless transport?
- A. Implement traffic shaping to smooth RDMA traffic bursts and prevent congestion.
- B. Set the RDMA traffic to use the highest DSCP value to ensure priority forwarding.
- C. Configure weighted fair queuing to prioritize RDMA traffic over other traffic classes.
- D. Enable Priority Flow Control on the traffic class carrying RoCEv2 packets.
Answer: D
Explanation:
RoCEv2 requires lossless Ethernet behavior to prevent packet drops that can disrupt RDMA communication. Enabling Priority Flow Control on the traffic class carrying RoCEv2 packets allows Ethernet pause behavior for that priority, helping maintain reliable, low-latency RDMA transport across the AI training fabric.
NEW QUESTION # 48
A server administrator must monitor the performance and utilization of GPUs in Cisco UCS servers running AI workloads. Which two metrics must be used in Cisco Intersight to accomplish these goals? (Choose two.)
- A. GPU Memory Utilization
- B. Monitor
- C. Active CPU Utilization
- D. Explorer
- E. Insights
Answer: A,C
Explanation:
Cisco Intersight performance monitoring uses resource utilization metrics to identify bottlenecks in AI workloads. Active CPU utilization shows whether host processing is limiting workload execution, while GPU memory utilization shows how effectively GPU memory is being consumed during model processing and training or inference tasks.
NEW QUESTION # 49
What is a purpose of Cisco AI PODs?
- A. to offer a flexible AI infrastructure focused on inferencing workloads with optional training capabilities
- B. to enable AI lifecycle management through cloud-native software with hardware support limited to compute servers
- C. to provide a turnkey, full-stack infrastructure for AI training, fine-tuning, and inferencing
- D. to deliver modular AI hardware components that can be integrated with third-party software platforms
Answer: C
Explanation:
Cisco AI PODs are designed as integrated, validated full-stack infrastructure solutions that combine compute, networking, storage, and management capabilities to support AI training, fine- tuning, and inferencing workloads with faster deployment and operational consistency.
NEW QUESTION # 50
A network architect designs a Cisco AI fabric with spine-leaf topology. What is the primary advantage of this architecture for AI training workloads?
- A. Simplified wireless integration
- B. Reduced need for IP addressing
- C. Elimination of routing protocols
- D. Predictable low-latency east-west traffic
Answer: D
Explanation:
AI workloads generate massive east-west traffic between compute nodes. Spine-leaf designs provide predictable latency and scalable bandwidth because each leaf switch maintains equal- cost paths to all spines. Routing protocols are still required, IP addressing remains necessary, and wireless integration is unrelated to AI fabric performance requirements.
NEW QUESTION # 51
How does RAG enhance the capabilities of LLMs?
- A. by providing access to proprietary data
- B. by retaining the LLM by altering its core parameters
- C. by removing inherent biases from LLM outputs
- D. by removing the data preparation and cleaning step
Answer: A
Explanation:
RAG enhances LLMs by retrieving relevant external or proprietary information at generation time and using that context to produce more accurate, current, and domain-specific responses without changing the model's core parameters.
NEW QUESTION # 52
An engineer must configure dynamic load balancing (DLB) on a Cisco Nexus 9000 Series Switch AI fabric using a CLI. Which configuration must be implemented for only the odd-numbered interfaces?
- A. Create a DLB hardware profile and include the list of desired interfaces.
- B. Exclude the desired interfaces from DLB with the interface level configuration of no dlb-interface.
- C. Under the relevant interface-level configuration, add the keyword dlb-interface.
- D. Create a QoS policy that enables DLB and apply that policy to the relevant interfaces.
Answer: C
Explanation:
Dynamic load balancing is enabled on the selected Cisco Nexus 9000 interfaces by applying the interface-level dlb-interface command. For this deployment requirement, the command is applied only to the odd-numbered interfaces that participate in the DLB configuration.
NEW QUESTION # 53
Drag and Drop Question
An organization is deploying an AI/ML infrastructure with multiple network fabrics, each serving a specific purpose. The network architect must match the network fabric type with its primary function. Drag and drop the network fabric types from the left onto the corresponding descriptions on the right.
Answer:
Explanation:
NEW QUESTION # 54
A data center is deploying an AI infrastructure for LLM training and model inference. This infrastructure requires high GPU density, low-latency interconnects, and the ability to scale compute resources. When considering the Cisco UCS C885A M8 Rack Server for the deployment, which design aspect is the most critical for achieving optimal performance and scalability in this environment?
- A. configuring the server with the minimum amount of RAM to reduce initial costs
- B. selecting the appropriate Nexus Fabric and its configuration to minimize latency
- C. implementing a redundant power supply configuration to ensure high availability
- D. using the maximum number of CPU cores to support the virtualization overhead
Answer: B
Explanation:
For UCS C885A M8 servers running LLM training and inference workloads, the surrounding Nexus fabric design is critical because GPU-dense systems depend on low-latency, high- bandwidth, lossless network connectivity for scale-out performance. Selecting and configuring the correct Nexus fabric minimizes communication delays between compute nodes and enables the AI infrastructure to scale efficiently.
NEW QUESTION # 55
A Cisco Nexus switch experiences head-of-line blocking in an RDMA environment. Which mechanism is most effective in reducing this condition?
- A. Virtual output queuing
- B. PortFast
- C. NAT overload
- D. Weighted Random Early Detection
Answer: A
Explanation:
Virtual output queuing (VOQ) prevents packets destined for congested outputs from blocking unrelated traffic flows. This is valuable in AI infrastructures where high-bandwidth GPU traffic can overwhelm buffers. WRED manages congestion probabilistically but does not eliminate head-of- line blocking. PortFast and NAT overload are unrelated features.
NEW QUESTION # 56
......
300-640 Test Engine files, 300-640 Dumps PDF: https://passitsure.itcertmagic.com/Cisco/real-300-640-exam-prep-dumps.html