🔥 3-day streak
NVIDIA-Certified Associate: AI Infrastructure and Operations81 / 145
Question 81 of 145

A cluster architect is designing the compute fabric for a large multi-node training deployment. During synchronized all-reduce operations, several links become hotspots and static routing sends multiple flows down the same path, causing congestion and stalling GPU communication. The architect wants a fabric feature that dynamically redistributes traffic across alternate paths to relieve these hotspots. Which InfiniBand capability should be enabled to address this?

Reviewed for accuracy · Report an issueNext question