300-640 DCAI: Implementing Cisco Data Center AI Infrastructure

By Cisco · Question Mode

QuestionQ23

AI Infrastructure Operations and Troubleshooting

A network operations team deploys HyperFabric AI and needs to monitor fabric health in order to proactively detect performance issues. The team must track metrics that are specifically relevant to AI training workloads. Which metric is most important for identifying congestion that could affect AI training performance in a HyperFabric AI deployment?

Explanation

Priority Flow Control (PFC) pause frames are sent when congestion reaches the configured buffer threshold for a lossless traffic class. A high PFC pause-frame transmission rate on an interface directly exposes congestion and backpressure that can slow distributed AI training communication. NVIDIA documents that PFC pause frames are generated after buffer consumption reaches the configured xoff threshold and that excessive pause frames indicate abnormal backpressure.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!