Snowflake has announced a new integration with NVIDIA, bringing GPU-accelerated CUDA-X Data Science libraries directly into the Snowflake ML environment. With NVIDIA’s cuML and cuDF libraries now preinstalled, data scientists can accelerate workloads such as Random Forest, UMAP, and HDBSCAN without modifying existing Python code. The integration is designed to streamline the full model-development lifecycle, enabling faster experimentation and reducing infrastructure overhead, according to Snowflake.
Christian Kleinerman, EVP of Product at Snowflake, stated, “By natively integrating NVIDIA CUDA-X libraries, we’re giving our customers a massive performance boost. And this isn’t just about faster performance; it’s about enabling our data scientists to spend less time on infrastructure and more time deriving insight.” According to NVIDIA benchmark tests, users can achieve performance gains up to 5x for Random Forest and as much as 200x for HDBSCAN when shifting from CPUs to NVIDIA A10 GPUs.
The collaboration expands the companies’ ongoing work to support advanced ML and generative AI within Snowflake’s platform. Pat Lee, VP of Strategic Enterprise Partnerships, NVIDIA, highlighted, “By integrating NVIDIA cuDF and cuML libraries directly into the Snowflake ML platform, customers can now harness accelerated computing with their existing Python workflows.” Through Snowflake’s Container Runtime, organizations can run demanding tasks such as large-scale topic modeling and computational genomics with significant time savings.