VAST Data has brought Nvidia GPU hardware and NIM microservices software into its AI storage and data processing to create an InsightEngine product providing real-time and automatically triggered AI model data access and analytical insights.
It has announced partnerships with Cisco and Equinix to widen its product’s route to market and set up a Cosmos AI user community centered on building an ecosystem of partners and users exchanging ideas around building AI deployments and use-cases using its products. We’ll cover the Cisco, Equinix and Cosmos news in separate stories and focus on the Nvidia GPU and NIM news here.
This VAST and Nvidia announcement builds on VAST’s storage+server data platform, which has a base of all-flash storage and a software stack comprising a Data Catalog, global namespace (DataSpace), unstructured DataStore, structured DataBase, and AI process-triggering Data Engine. Its all-flash storage has a DASE (Disaggregated and Shared-Everything Architecture) with scale-out x86-based controller nodes (C-nodes) linking to data-storing, all-flash, D-nodes across InfiniBand or RoCE links with a 200/400 Gbps networking fabric. The C-node and D-node software can run in shared industry-standard servers and the D-node software can also run in Nvidia BlueField3 DPUs.
The existing Nvidia partnership has VAST’s system certified for the DGX SuperPOD. This has been extended so that Nvidia GPUs can now be VAST controller nodes:
That means that the GPUs can work directly on data stored in the VAST array without it having to be first moved to the GPU server. Secondly, Nvidia’s NIM microservices software now runs natively inside the VAST software environment. NIM provides Gen AI Large language models (LLMs) as optimized containers. These simplify and accelerate the deployment of custom and pre-trained AI models across clouds, datacenters and workstations.
Justin Boitano, VP Enterprise AI at Nvidia, stated: “Integrating Nvidia NIM into VAST InsightEngine with Nvidia helps enterprises more securely and efficiently access data at any scale to quickly convert it into actionable insights.”
VAST says its software with NIM embeds “the semantic meaning of incoming data using advanced models powered by Nvidia GPUs. The vector and graph embeddings are then stored in the VAST DataBase within milliseconds after the data is captured to ensure that any new file, object, table or streaming data* is instantly ready for advanced AI retrieval and inference operations.”
InsightEngine uses VAST’s DataEngine to trigger the Nvidia NIM embedding agent as soon as new data is written to the system, allowing for real-time creation of vector embeddings or graph relationships from unstructured data. Such vectors and graphs are used in RAG (Retrieval-Augmented Generation) whereby a customer’s proprietary data is used to inform LLM query responses to make them more accurate and less prone to fabricate data relationships (hallucinations).
The VAST DataBase can store “exabytes of both structured and unstructured enterprise datasets” and “trillions of embeddings,” and run “real-time similarity search across massive vector spaces and knowledge graphs.”
Data indexing occurs at the point of data ingestion and “this architecture eliminates the need for separate data lakes and external SaaS platforms.” The data is held and processed securely. VAST says “any file system or object storage data update is atomically synced with the vector database and its indices, offering comprehensive, secure data access management and global data provenance to ensure data consistency across multi-tenant environments.”
Jeff Denworth, co-founder at VAST Data, stated: “With the VAST Data Platform’s unique architecture, embedded with Nvidia NIM, we’re making it simple for organizations to extract insights from their data in real-time. By unifying all elements of the AI retrieval pipeline into an enterprise data foundation, VAST Data InsightEngine with Nvidia is the industry’s first solution to provide a universal view into all of an enterprise’s structured and unstructured data to achieve advanced AI-enabled decision-making.”
VAST’s InsightEngine with Nvidia will be generally available in early 2025. Learn more here.
Bootnote
*We understand block data support is coming to VAST Data.
