NetApp launches Novus storage architecture for AI workloads

NetApp, a leading storage technology company, has unveiled Novus, a new storage architecture tailored for AI infrastructure providers managing extensive GPU environments. As reported by NetApp, the innovative architecture separates metadata from the data path, allowing performance, capacity, and concurrency to scale independently under a unified namespace using standard NFS.

The initial version of Novus integrates NetApp Novus Data Director metadata software running on Supermicro hardware with NetApp ONTAP data services through AFF A90 systems. According to the company, the architecture is designed to support data transfer rates surpassing 100 terabytes per second and accommodate high usage of hundreds of thousands of GPUs.

A statement from Syam Nair, NetApp's Chief Product Officer, emphasizes the necessity of AI factories having an architecture capable of handling data demands, as HPC-era storage alone cannot meet the needs of AI Factory scale. NetApp's architecture promises to support billions of files and datasets in a single namespace, enabling the independent scaling of metadata, bandwidth, and capacity.

This approach aims to minimize storage-related bottlenecks in multi-tenant, cloud-scale AI environments, according to NetApp. Omdia, an independent research firm, validated NetApp Novus by observing near-linear performance scaling as ONTAP clusters were added to a single global namespace. Omdia's modeling projects the architecture to achieve near 100 terabytes per second sequential read throughput with hundreds of exabytes of effective capacity.

It is important to note that details about unreleased offerings and future plans are subject to change without prior notice and should not be considered as a basis for making purchasing decisions.