As part of the full-stack NVIDIA Vera Rubin platform, NVIDIA Minerva turns full-stack telemetry from compute and NVLink switching into proactive maintenance, agentic debugging, and closed-loop remediation. This protects GPU hours, cluster uptime, and token production to maximize AI factory yield.
NVIDIA Minerva
The AI-native AIOps platform for maximizing AI factory yield.
Overview
Transforming AI Factory Operations
NVIDIA Minerva™ is the primary observability solution for NVIDIA AI factories. An AI-native, end-to-end AIOps platform, it’s purpose-built to maximize the availability and performance of the AI factory infrastructure stack.
As part of the full-stack NVIDIA Vera Rubin platform, NVIDIA Minerva ingests telemetry from NVIDIA GPUs, Spectrum-X™ Ethernet, NVLink™ switches, SuperNICs, DPUs, schedulers (such as Slurm and NVIDIA Run:ai), and AI workloads. A dynamic knowledge graph, machine learning (ML), and agentic AI deliver real-time anomaly detection, automated root cause analysis, and closed-loop remediation. AI factories move from reactive to proactive maintenance with faster agentic debugging and a reduction in alert noise.
AIOps Observability Solution for NVIDIA AI Factories
NVIDIA Minerva correlates telemetry across NVIDIA compute, networking, and full-stack AI software offerings.
NVIDIA AI Factory
Maximize the resiliency and performance of the NVIDIA Vera Rubin platform.
NVIDIA DSX OS
Integrate observability into operations and management of the DSX stack.
NVIDIA Spectrum-X Ethernet
Harness high-frequency telemetry for end-to-end network observability.
Features and Benefits
Higher Infrastructure Utilization for More AI Throughput
Keep the AI factory running at full efficiency to maximize yield, accelerate model training and inference, and increase token production.
Maximized Compute Uptime
Actionable intelligence recovers GPU hours that would otherwise be lost to undetected failures.
From Reaction to Prevention
Emerging failures and performance degradations are caught before they take down training jobs.
Reduced Operational Noise
Advanced noise reduction turns event storms into actionable alerts.
Flexible Deployment
Multiple deployment options from DIY to full turnkey adapt to your organization.
Predictive Maintenance
Automated predictive maintenance increases mean time between interrupts (MTBI).
Root Cause Analysis
Root cause analysis of stragglers and failed jobs accelerates recovery and reduces mean time to repair (MTTR).
Unified View
Real-time telemetry across hardware, schedulers, and workloads provides a single pane of glass to diagnose problems.
More GPU Hours
More effective GPU hours lift token production and cut time to train.
Capabilities
Full-Stack Intelligence for the AI Factory
Correlate telemetry across infrastructure and workloads in one dynamic, real-time view.
Product
How NVIDIA Minerva Works
Collect telemetry from every layer of an AI factory, and apply ML anomaly detection and correlation to achieve actionable intelligence. Data and insights are integrated with NVIDIA Minerva’s UI and AI agent, as well as with external observability platforms and agents.
Use Cases
AIOps Purpose-Built for AI Factories
Validate infrastructure, investigate job and cluster issues, reduce noise, and prevent disruptions before they affect output.
-
Proactive Maintenance
-
Agentic Debugging
-
Full-Stack Correlation
-
Noise Reduction
-
Network Observability
Proactive Maintenance
NVIDIA Minerva uses real-time, ML-based anomaly detection to identify infrastructure issues before they affect production. By continuously monitoring the full AI factory stack and distilling billions of telemetry data points into high-fidelity, actionable alerts, NVIDIA Minerva gives operations teams the advance warning they need to act with confidence.
This shift from reactive troubleshooting to proactive maintenance reduces MTTR, preventing hundreds of lost GPU hours and keeping AI workloads running at peak performance.
Agentic Debugging
NVIDIA Minerva enables agentic, topology-wide root cause analysis for AI factories. It has both a native AI debugging assistant called Sherlock and MCP-based integration with other AI agents such as OpenClaw. Together, these capabilities fit seamlessly into existing operational workflows.
Instead of manually sifting through logs and dashboards, operators can ask questions in natural language and receive direct, actionable answers about what’s happening across the AI factory.
Full-Stack Correlation
AI factories span an enormous range of interconnected systems, from GPU nodes and NVLink fabric to Spectrum-X Ethernet, job schedulers, workload managers such as Slurm, and active workloads. When an issue occurs, its root cause may originate in one layer but appear as a symptom in another.
By ingesting full-stack telemetry in real time and applying ML-based, topology-aware correlation, NVIDIA Minerva reveals the relationships that matter: how a network anomaly affects GPU utilization or how a scheduler event contributes to degraded job throughput.
Operations teams gain a single, unified view of the AI factory instead of the fragmented, siloed tools that make cross-layer debugging slow and costly.
Noise Reduction
Alert fatigue is one of the most underappreciated costs in AI factory operations. A network issue that cascades across hundreds of components may generate thousands of alerts. NVIDIA Minerva uses topology-aware correlation to map AI factory dependencies and consolidate related events into a single, contextualized incident. NVIDIA Minerva distills these alerts into a single, precise, prioritized signal with actionable context delivered through tools such as Slack or ServiceNow.
Teams spend less time triaging alerts, diagnose issues faster, and focus on the highest-impact operational problems.
Network Observability
NVIDIA Minerva is the primary network observability solution for NVIDIA AI factories. NVIDIA Minerva includes out-of-the-box dashboards and alerts purpose-built for Spectrum-X Ethernet networks, NVLink scale-up fabric, hardware health, and services. Dashboards are aligned with NVIDIA reference architectures, giving operations teams best-practice visibility.
Integration with DSX OS extends this capability further, enabling NVIDIA Minerva to share components, data models, and operational context across the broader DSX ecosystem.
Consumption Models
Deploy NVIDIA Minerva With Flexible Consumption Models
For NVIDIA Minerva’s data collector, real-time anomaly detection, and AI capabilities, choose the configurations, from DIY to full turnkey, best suited to your organization’s workflows.
Resources
Learn More About NVIDIA Minerva
NVIDIA Minerva FAQs
NVIDIA Minerva is an AI-native AIOps platform that correlates infrastructure and workload telemetry to provide full-stack observability across compute, networking, storage, and AI workloads and converts these into insights for proactive or automated action.
NVIDIA Minerva gathers data from NVIDIA infrastructure, network, schedulers, and workload telemetry. This full-stack context allows it to correlate anomalous interfaces with AI jobs, helping operators isolate root causes that generic tools and siloed point solutions may miss.
NVIDIA Minerva is the primary network observability solution for NVIDIA AI factories, NVIDIA Vera Rubin, and NVIDIA Blackwell. It spans NVIDIA Spectrum-X Ethernet for scale-out, scale-across, and scale-in, NVIDIA NVLink for scale-up, Slurm and other schedulers, and AI workloads.
NVIDIA Minerva can be deployed on premises or consumed as SaaS from NVIDIA cloud. It supports multiple consumption models from DIY to full turnkey solutions to fit existing AI factory workflows.
NVIDIA Minerva ingests compute, network, storage, scheduler, NCCL, and workload telemetry through standards, including OpenTelemetry, gNMI, syslog, DCGM, Slurm, and JSON. It integrates with Slack, Microsoft Teams, PagerDuty, ServiceNow, Grafana, Ansible, and infrastructure APIs.
NVIDIA Minerva uses purpose-built AI/ML to learn normal behavior, predict anomalies, correlate events, and recommend remediation. A native AI assistant enables natural-language triage using live telemetry, while its MCP server connects NVIDIA Minerva with third-party agentic tools.
NVIDIA Minerva reuses DSX OS components, while key NVIDIA Minerva capabilities, starting with NVIDIA Minerva Collector, will be open sourced in DSX OS, creating a modular observability foundation.
Next Steps