Snowflake unveils AI, open data, and dev advancements at Summit 2024: NVIDIA collab, Polaris Catalog for Iceberg, Cortex AI upgrades, Snowflake ML, and more.
At its annual Snowflake Summit 2024 conference, Snowflake made several significant announcements advancing its offerings in AI, open data interoperability, and developer experience on its unified platform.
New Collaboration with NVIDIA for Customized Enterprise AI Applications
Snowflake is collaborating with NVIDIA to allow customers and partners to build customized AI data applications in Snowflake powered by NVIDIA AI. This includes:
Integrating NVIDIA AI Enterprise software like NeMo Retriever microservices into Snowflake Cortex AI for highly accurate applications
Full support for Snowflake Arctic large language model (LLM) with NVIDIA TensorRT-LLM software for optimized performance
Making Snowflake Arctic available as an NVIDIA NIM inference microservice for efficient intelligence
This allows enterprises to create bespoke AI solutions using their business data rapidly.
Polaris Catalog for Open Interoperability with Apache Iceberg
Snowflake unveiled Polaris Catalog, an open, vendor-neutral catalog implementation for Apache Iceberg that provides complete enterprise security and interoperability with AWS, Google Cloud, Microsoft Azure, Salesforce and more. Polaris Catalog can be hosted or self-hosted in Snowflake's AI Data Cloud. It relies on Iceberg's open REST protocol to allow any engine supporting the Iceberg REST API to find and access an organization's Iceberg tables.
New Cortex AI Innovations for Easy, Efficient, Trusted Enterprise AI
Snowflake announced several advancements to Snowflake Cortex AI:
New chat experiences Cortex Analyst and Cortex Search allow the development of custom chatbots in minutes against structured and unstructured data
Cortex Guard leverages Meta's Llama Guard to filter harmful content
Document AI extracts content from documents using Snowflake Arctic-TILT multimodal LLM
Snowflake Copilot combines Mistral Large with a proprietary model to accelerate SQL development
No-code Snowflake AI & ML Studio interface to develop and productize AI apps
Cortex Fine-Tuning for customization of Meta and Mistral AI models
Streamlined Model Development and Operations with Snowflake ML
Snowflake ML brings MLOps capabilities natively into the AI Data Cloud:
Model Registry to govern model access and use
Feature Store to create, manage, and serve ML features
ML Lineage to trace feature, dataset, and model usage
Open Data Interoperability and Collaboration
In addition to Polaris Catalog, Snowflake made its open data support and collaboration capabilities easier:
Iceberg Tables GA for using Snowflake's governance, performance on external Iceberg data
Internal Marketplace to curate data products for internal teams
Sharing of AI models, Iceberg Tables, Dynamic Tables
Universal Search to find content across Snowflake, external storage, 3rd parties
AI-powered object descriptions to aid data discovery
Enhanced Platform Performance and Efficiency
Snowflake continues to enhance core platform price/performance:
27% reduction in query duration since tracking started, 12% in last 12 months
Up to 25% and 50% faster loading of JSON and Parquet data, respectively
Expansion to new regulated and DoD environments
Accelerated App and Model Development for Builders
Snowflake introduced new tools to speed the development of pipelines, models, and apps:
Snowflake Notebooks for single interface Python/SQL development
Snowpark pandas API for familiar pandas operations at Snowflake scale
Database Change Management and Git integration for declarative DevOps
Snowflake Trail observability for monitoring data quality and pipelines
Native App Framework integration with Snowpark Container Services for GPU-powered apps
These announcements demonstrate Snowflake's commitment to being the leading provider of a unified platform for enterprise AI, open data interoperability, and accelerated development. The company continues to rapidly innovate to help organizations extract maximum value from their data assets while maintaining strong governance and efficiency.
Comments