VAST Data, an AI data platform company, has introduced an innovative AI cloud architecture designed to revolutionize performance, security, and efficiency in AI operations. Developed in partnership with NVIDIA and deployed at CoreWeave, this groundbreaking infrastructure leverages NVIDIA BlueField DPUs to streamline and scale AI processes.
The new architecture marks a significant advancement in AI computing, leveraging NVIDIA BlueField-3 DPUs to integrate VAST’s parallel data services architecture directly into AI servers. This integration allows for the seamless disaggregation of VAST’s operating system into AI computing machinery, effectively transforming conventional supercomputers into AI data engines.
By harnessing the power of the NVIDIA BlueField networking platform, which combines robust compute capabilities with integrated hardware accelerators, the new architecture delivers unparalleled performance and security for AI operations. Each GPU server is equipped with a dedicated NVIDIA BlueField DPU, running a stateless container to power VAST’s parallel services operating system. This design embeds storage and database processing services directly into AI servers, enabling true linear data services that can scale to accommodate hundreds of thousands of GPUs.
Moreover, the architecture streamlines infrastructure by eliminating multiple layers of x86 hardware and networking, resulting in significant cost, footprint, and power savings for AI data services.
Key benefits of the collaboration with NVIDIA include:
– Maximizing Data Center Efficiency: VAST’s Disaggregated, Shared Everything (DASE) architecture, powered by NVIDIA BlueField-3 DPUs, reduces the power usage and data center footprint by 70%, leading to a net energy consumption savings of over 5% compared to previous infrastructure.
– Ensuring Unprecedented Quality of Service: Each GPU server is equipped with a dedicated parallel storage and database container, eliminating contention for data services infrastructure and ensuring optimal performance for multi-tenant environments.
– Enhancing Zero-Trust Security: The architecture isolates data and data management from host operating systems, minimizing attack vectors and ensuring robust security for AI operations.
– Delivering Block Storage Services: VAST systems, powered by the NVIDIA DOCA software framework, now provide block storage services natively to host operating systems, enhancing the versatility of data presentations for high-performance applications.
Jeff Denworth, co-founder at VAST Data, expressed pride in the partnership with NVIDIA, highlighting the architectural innovation enabled by NVIDIA BlueField-3 DPUs.
The new VAST architecture is currently undergoing testing and deployment at CoreWeave, a leading GPU cloud provider. Peter Salanki, vice president of Engineering at CoreWeave, emphasized the transformative impact of the architecture on data center operations and user experience.
Rob Davis, Vice President of Storage Technology at NVIDIA, praised VAST’s revolutionary architecture for its ability to simplify and secure AI computing experiences.
The collaboration between VAST Data and NVIDIA represents a significant step forward in the evolution of AI infrastructure, offering unprecedented levels of performance, security, and efficiency for AI operations.
