Nvidia launches Minerva to catch failing parts in giant AI data centres (thenextweb.com)

Nvidia has introduced Minerva, an AI operations software designed to monitor large AI data centres and identify failing components before they disrupt operations. Launched at the GTC Berlin conference, Minerva aggregates data from GPUs, networks, storage, and workloads, allowing operators to trace performance issues back to specific parts of the system, thereby preventing costly delays. This unified approach addresses a common challenge in massive data centres, where disparate teams often overlook critical signs of failure. Significantly, Minerva employs machine learning to establish a baseline of normal operations, enabling it to predict potential failures and consolidate alerts into single incidents with suggested causes. Its built-in AI assistant allows operators to inquire about cluster health in simple language and receive actionable suggestions. Notably, prominent companies like IREN, Lambda, IBM, and Tesla plan to implement Minerva, emphasizing its potential impact on the efficiency and reliability of AI operations. The launch reflects Nvidia's broader strategy to diversify its offerings beyond GPUs, aiming to enhance operational sustainability in AI data centres while also promoting industry partnerships to improve resource usage. Pricing details for Minerva have yet to be disclosed.