The NVLink dashboards cover the high-speed, point-to-point interconnects between GPUs within a node. They monitor east-west traffic for collective operations like All-Reduce and track chip-to-chip communication bandwidth.
-
NVLink Health Dashboard: Monitors the operational status of NVLink interconnects, including link-down events and hardware signal degradation.
-
NVLink Interfaces: Shows per-interface bandwidth utilization and error counters to help debug performance jitter during training workloads.
Last updated: