Nvidia launches Minerva software to detect AI data center faults
IREN is using the monitoring system, while IBM, Lambda and Tesla are among companies planning to adopt it. Nvidia has not disclosed pricing.
Nvidia introduced Minerva at its GTC Berlin conference, saying the software can identify potential equipment failures in large AI computing clusters before they disrupt workloads. The system brings together monitoring information that would otherwise be split among teams responsible for different parts of a data center.
According to Nvidia, Minerva combines information from graphics processors, networking equipment, storage, scheduling systems and AI workloads. The company says operators can use that combined view to locate the hardware behind a slow-running job, including in installations containing hundreds of thousands of GPUs.
Machine learning establishes normal operating patterns and predicts failures, Nvidia said. Related alerts are consolidated into incidents with a probable cause. An integrated AI assistant answers staff questions in everyday language and recommends remedies, which operators can implement manually or through their existing automation systems.
IREN already uses Minerva. Nvidia said planned users also include Tesla, Jio, IBM, Bitdeer AI and Lambda, while Supermicro, Dell and Lenovo will make it available to customers. The launch extends Nvidia’s offerings beyond chips for AI data centers. The company did not disclose Minerva’s price.
