NVIDIA Unveils Minerva to Bring End-to-End Observability for AI Factories
nvidia ·
NVIDIA has launched Minerva, an AIOps platform designed to provide end-to-end observability for complex AI factories. The system integrates telemetry from infrastructure components like GPUs and networking with workload data to create a dynamic knowledge graph. By applying purpose-built AI models, Minerva analyzes billions of data points to detect anomalies, predict failures, and correlate events based on factory topology. This approach allows operators to trace performance issues across hundreds of thousands of GPUs and group alert storms into focused incidents with likely root causes. The platform features a native AI assistant that offers natural-language interaction and recommended remediation steps to support proactive maintenance. Key benefits include extended time between interruptions, faster repair times, and improved insight by filtering out irrelevant data. Major partners such as IREN, IBM, Tesla, and hardware vendors like Dell Technologies are already adopting or offering the solution. These capabilities aim to maximize GPU uptime and overall yield for large-scale AI operations.
nvidia