NetApp-Nvidia partner to add RAG capabilities to OnTap
NetApp OnTap now enables data stored in the cloud and on premises to connect with LLMs via Nvidia's NeMo RAG -- a feature that will likely become table stakes for storage vendors.
Customers of NetApp's OnTap operating system can now use their existing data for AI app creation through a new Nvidia cloud capability.
The Nvidia NeMo Retriever microservice, a retrieval-augmented generation (RAG) software offering, now connects directly to OnTap customer hybrid-cloud storage environments thanks to a new partnership the vendors unveiled at this week's Nvidia GTC 2024 conference.
This feature is available to OnTap customers that subscribe to the Nvidia AI Enterprise SaaS and maintains OnTap's data security and privacy controls to prevent misuse in AI projects.
Giving a large language model (LLM) access to an enterprise's unstructured data without having to create a separate repository could be useful to get an AI project off the ground, said Dave Raffo, an independent storage analyst.
However, he doesn't expect Nvidia to play favorites with storage vendors and sees similar capabilities coming to other storage OSes. In fact, Pure Storage, another hybrid cloud storage vendor specializing in flash media, also released a similar feature and partnership at GTC using NeMo.
"It's certainly helpful to NetApp to customers, but I don't know how specific it will be," Raffo said. "I wouldn't be surprised if other [storage vendors] will connect as well."
Finding for NeMo
Nvidia NeMo is a cloud framework to build, customize and deploy generative AI (GenAI) models, with Nvidia AI Enterprise providing additional support, reference architectures and management tools over open source models.
RAG enables enterprises to create a customized data set using proprietary information without the LLM ingesting that data, allowing an enterprise to use GenAI without opening itself up to risks such as leaks.
It's another example of AI as an enabling technology, [but NetApp is] not really enhancing the storage.
Dave RaffoIndependent storage analyst
NetApp customers can now use the additional data control and security features in OnTap to run unstructured data anywhere it rests in the enterprise through a RAG, according to Andy Sayare, director of global strategic alliances for AI at NetApp.
"You have no concerns over security and privacy because you're not pushing this data into the model," Sayare said. "[NetApp] has all kinds of security and access control in the data and that is all preserved when that goes to the large language model."
Having such a capability is useful as teams may need to quickly create prototype AI applications, Raffo said. But none of this adds to the core storage technology NetApp sells.
"It's another example of AI as an enabling technology, [but NetApp is] not really enhancing the storage," he said.
Retrieval agitation generation
Enterprise storage teams will remain focused on providing the infrastructure backbone for data rather than specific GenAI tools or capabilities promised by vendors, according to Marc Staimer, founder and president of Dragon Slayer Consulting.
Using RAGs with an LMM is safer and more effective than direct training and can help to prevent data leaks or hallucinations, he said. But creating vector databases and other components still requires customers to consider who to trust with their proprietary data, how that data is used in the LLM and what controls exist to curb misuse.
"If you don't have to move the data, that's a good thing," Staimer said. "The question becomes, what are [Nvidia and NetApp] doing? Can you choose accuracy over speed or speed over accuracy? Do the users even have that choice?"
Tim McCarthy is a news writer for TechTarget Editorial covering cloud and data storage.
