Sferical AI - NVIDIA DGX™ GB300
NVIDIA DGX™ GB300 is a comprehensive AI infrastructure solution designed for the era of AI reasoning and uniquely suited for training and inference of state-of-the-art models. DGX GB300 is built with Grace Blackwell Ultra Superchips and can be scaled up to tens of thousands of Superchips with NVIDIA DGX SuperPOD, creating a massive shared memory space that accelerates the performance of the world’s largest AI models. Its rack-scale, 100% liquid-cooled design is tailored to fit into contemporary data centers using NVIDIA MGX™ racks, helping ease the burden on enterprises to manage high-performance AI hardware. DGX GB300 equips organizations with an infrastructure solution that can break through performance bottlenecks and handle the most demanding AI workloads.

Built on NVIDIA Grace Blackwell Ultra
DGX GB300 represents a transformative leap in AI computing capabilities, powered by the NVIDIA Grace Blackwell Ultra architecture. This leading-edge system to delivers unprecedented inference performance while enhancing training capabilities across any AI workload. Powered by 36 Grace CPUs and 72 Blackwell Ultra GPUs that are connected into one massive GPU through an NVIDIA NVLink Switch System, DGX GB300 delivers 1.4 exaFLOPS for inference and 360 PFLOPS for training. DGX GB300 is purpose-built to optimize performance for the entire AI pipeline, from model training and post-training optimization to test-time inference, enabling enterprises to scale their infrastructure to meet the demands of the era of AI reasoning.

Run Models, Automate the Essentials
With NVIDIA Mission Control
NVIDIA Mission Control powers every aspect of AI factory operations, fromdeveloper workloads to infrastructure to facilities, with the skills of a world-class operations team, now delivered as software. It brings instant agility for inference and training while providing full-stack intelligence for infrastructure resilience. Mission Control lets every enterprise run AI with hyperscale-grade efficiency, accelerating AI experimentation. Additionally, NVIDIA AI Enterprise, offering a suite of software to streamline AI development and deployment, is optimized to run on NVIDIA DGX systems. Use NVIDIA NIM™ microservices for optimal model deployment, offering speed, ease of use, manageability, and security.




Designed for the Modern Data Center
DGX GB300 is engineered to seamlessly integrate with modern data center environments, offering unparalleled flexibility and scalability. With DGX GB300, enterprises are empowered to operate with hyperscaler-level capabilities without the need to overhaul their existing infrastructure. With a 100% liquid-cooled rack design, DGX GB300 significantly enhances energy efficiency and allows for higher power density crucial for advanced AI workloads. By combining hyperscaler-level performance with the adaptability to work within existing infrastructures, DGX GB300 represents a transformative solution for organizations seeking to harness the full potential of AI without compromising on efficiency or compatibility.