NVIDIA Minerva: Innovation for AI Factories

NVIDIA has today launched Minerva, an advanced AIOps platform designed to enable teams to detect issues earlier and ensure the productivity of AI factories. Minerva collects telemetry from both infrastructure and AI workloads and applies specially developed AI models to convert this data into timely, actionable insights. With the growing complexity of AI factories, where countless GPUs work together over high-speed networks, even the smallest anomalies, such as an overheated GPU, can lead to significant delays in the overall process.

The focus of Minerva is on the yield of AI factories, a measure of the usable output produced by the infrastructure, such as GPU hours and cluster uptime. This provides a comprehensive view of the operational status. Unlike traditional monitoring tools, which often look at data centers from various perspectives, Minerva correlates signals from NVIDIA GPUs, networks, and storage together. This creates a dynamic knowledge graph that maps the whole and allows operators to quickly trace issues in complex environments.

Minerva processes billions of data points per hour through both supervised and unsupervised machine learning. However, the system not only learns to recognize normal behavior, but also detects unforeseen anomalies and predicts potential failures. This simplifies operations, as Minerva aggregates related issues into a few actionable items, allowing AIOps teams to shift from reactive to proactive maintenance. All of this contributes to maximizing the uptime of GPUs, which is crucial for the effectiveness of AI factories.

IREN, one of NVIDIA's partners, has already announced plans to deploy Minerva for optimizing their AI factories. According to Denis Skrinnikoff, CTO of IREN, Minerva provides insight into the performance of network, GPUs, and workloads, allowing their team to focus on proactive maintenance rather than troubleshooting. In addition to IREN, Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI, and Tesla also have plans to roll out Minerva. Key hardware partners such as Dell Technologies, Lenovo, and Supermicro will also integrate Minerva into their AI factory portfolio.

Read the full article from Nvidia.