At Monday’s Snowflake user conference in Las Vegas, Snowflake Summit 2023, the cloud database maker announced a partnership with chip giant Nvidia.
Under the deal, Snowflake customers will be able to rent GPU capacity from Snowflake’s data warehouses, and use that capacity to fine-tune neural networks with Nvidia’s NeMo framework, announced last fall. Basic models are very large neural networks, such as LLMs, which are usually “pre-trained”, that is, they have already been developed to a certain level of capability.
Concretely, a customer will use Snowflake’s data warehouse to use their own data to develop a customized version of the NeMo base model to meet their needs.
Nvidia
Snowflake will implement the service by purchasing Nvidia GPU instances from cloud service providers
“It’s a very natural combination for both companies,” Nvidia vice president Manuvir Das said at a press conference. Mr. Das continued:
Nvidia and Snowflake figure that if companies need to build custom models for generative AI from their data, and that data is in Snowflake’s cloud, they need to integrate Nvidia’s modeling engine, NeMo. , in the Snowflake data cloud. This will allow enterprise customers to produce these models, directly in their data cloud, which they can then use for their business application cases.
This announcement is part of the growing development of the use of generative AI. On Monday, Apache Spark developer Databricks stunned the tech industry with a $1.3 billion acquisition of startup MosaicML, which runs an AI model training and deployment service.
Snowflake will implement the service by purchasing Nvidia GPU instances from the cloud service providers it already works with. “We are now talking about extending this relationship to include GPU-based instances,” Das said.
Nvidia has made software a major driver of growth for its business
In response to a question from ZDNET about how customer data would be protected, Manuvir Das said the primary responsibility lies with Snowflake.
“Snowflake has a design that ensures that when a customer chooses to perform computations on the Snowflake data cloud, it stays within that customer’s boundaries, Das said, and the NeMo engine fits into that model. He added, “Certainly NeMo also has a responsibility ‘in terms of security,’ and that’s why it’s a joint engineering job.”
This partnership follows Nvidia’s recent announcement with ServiceNow to use NeMo with ServiceNow customers. While the Snowflake deal is “general purpose,” Das said, the partnership with ServiceNow “is more of an ISV (independent software vendor) type model.” ServiceNow uses NeMo to train customer models for each of its customers, “so that when each of its customers opens tickets, they get specific responses.”
Nvidia CEO Jensen Huang has made software a major driver of growth for his company, which earns billions selling GPU hardware to develop neural networks. NeMo is part of the enterprise software stack that the company promotes, largely through partnerships with cloud providers.
In March, Colette Kress, Nvidia’s chief financial officer, told investors at a Morgan Stanley conference: “Our software business currently represents hundreds of millions of dollars in revenue and we still consider it to be a growth opportunity for the future”.
Source: “ZDNet.com”