High-severity Nvidia bug could crash GPU monitoring on exposed servers

Researchers found thousands of GPU servers exposing Nvidia DCGM Exporter and Prometheus/Node Exporter metrics to the public internet; a high-severity bug (tracked by Nvidia and rated CVSS 8.2) in DCGM Exporter could let unauthenticated requests exhaust memory and crash the exporter, disrupting GPU monitoring and potentially impacting AI training/inference. The scans identified ~2,100 exposed DCGM hosts (≈12,000 GPU UUIDs) across ~300 organizations and ~12,096 public Node Exporter hosts; operators are advised to restrict access and upgrade DCGM Exporter to version 4.8.2 or later.