NVIDIA Unveils Minerva to Bring End-to-End Observability for AI Factories
NVIDIA introduced Minerva, an AIOps platform designed to enhance observability and reliability in AI factories by integrating telemetry from infrastructure and AI workloads.
NVIDIA Minerva is a newly launched AIOps platform aimed at improving the productivity of AI factories by identifying issues earlier. It consolidates telemetry from infrastructure and AI workloads, using purpose-built AI models to generate actionable insights. AI factories rely on vast numbers of GPUs and high-speed networks, where even minor component degradation can disrupt entire operations. Minerva focuses on metrics like GPU hours, goodput, cluster uptime, and token production to measure infrastructure effectiveness.
The platform addresses the challenge of fragmented monitoring tools that isolate data center teams by connecting signals across GPUs, networking, storage, schedulers, and workloads. Minerva constructs a dynamic knowledge graph that maps infrastructure components such as GPUs, SuperNICs, switches, and cables alongside running jobs. This mapping enables operators to trace slow training jobs to specific components, even in factories with hundreds of thousands of GPUs.
Minerva processes billions of data points hourly using supervised and unsupervised machine learning to detect anomalies, predict failures, and correlate events based on factory topology. Instead of overwhelming operators with individual symptoms, it groups related issues into focused incidents with likely root causes. The platform includes a native AI assistant and integrates with external agentic AI tools, offering a natural-language interface for investigating cluster health and receiving remediation recommendations.
The platform’s ability to correlate workload and infrastructure issues, combined with its agentic AI capabilities, simplifies AI factory operations by reducing alert overload. This shift from reactive to proactive maintenance helps maximize GPU uptime. NVIDIA partners like IREN, Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI, and Tesla plan to deploy Minerva, with Dell Technologies, Lenovo, and Supermicro offering it as part of their AI factory operations portfolios.