SANTA CLARA — In a landmark announcement, NVIDIA has introduced the cutting-edge DGX SuperPOD powered by the formidable Grace Blackwell Superchips, heralding a new era in AI supercomputing. This groundbreaking development was unveiled today at the prestigious GTC event, showcasing NVIDIA’s relentless commitment to advancing the frontier of artificial intelligence.
The DGX SuperPOD represents a monumental leap forward in processing capability, offering unparalleled performance for handling trillion-parameter models with unwavering uptime, ideal for superscale generative AI training and inference tasks. Boasting a liquid-cooled rack-scale architecture, this next-generation AI supercomputer is a testament to NVIDIA’s dedication to pushing the boundaries of technological innovation.
At the heart of the DGX SuperPOD lies the revolutionary NVIDIA DGX GB200 systems, equipped with 36 NVIDIA GB200 Superchips each, comprising 36 Grace CPUs and 72 Blackwell GPUs. These Superchips, interconnected via fifth-generation NVIDIA NVLink®, deliver an astounding 30x performance boost for large language model inference workloads compared to previous iterations, empowering users with unprecedented computational prowess.
Jensen Huang, the visionary founder and CEO of NVIDIA, emphasized the transformative potential of the DGX SuperPOD, describing it as the cornerstone of the AI industrial revolution. With its ability to scale to tens of thousands of GB200 Superchips interconnected via NVIDIA Quantum InfiniBand, the DGX SuperPOD offers a glimpse into the future of AI infrastructure, enabling organizations to refine and generate their own AI solutions at an unprecedented scale.
The architectural marvel of the DGX SuperPOD extends beyond raw processing power, featuring a unified compute fabric incorporating advanced technologies such as NVIDIA BlueField®-3 DPUs and Quantum-X800 InfiniBand networking. This innovative design, coupled with fourth-generation NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP)™ technology, ensures unparalleled bandwidth and computational efficiency, setting a new standard for generative AI architectures.
Furthermore, the DGX SuperPOD offers a turnkey solution that seamlessly integrates with high-performance storage solutions from NVIDIA-certified partners, streamlining deployment and maximizing uptime. With intelligent predictive-management capabilities, the DGX SuperPOD continuously monitors thousands of data points to preemptively identify and address potential sources of downtime, ensuring uninterrupted operation and optimal resource utilization.
In addition to the groundbreaking DGX SuperPOD, NVIDIA also unveiled the DGX B200 system, a versatile AI supercomputing platform tailored for AI model training, fine-tuning, and inference tasks. Featuring the sixth-generation Blackwell architecture, the DGX B200 system delivers unmatched performance and scalability, making it an indispensable asset for industries worldwide.
As organizations embark on their AI journey, NVIDIA remains steadfast in its commitment to providing comprehensive software solutions and expert support to facilitate seamless integration and deployment. With NVIDIA AI Enterprise software and the expertise of certified partners, customers can harness the full potential of NVIDIA DGX platforms, accelerating their path to AI-driven innovation.
The DGX SuperPOD with DGX GB200 and DGX B200 systems is slated for release later this year, marking a significant milestone in the evolution of AI supercomputing. For more information, attendees are encouraged to visit the NVIDIA booth at GTC or watch a replay of the keynote address, where the future of AI infrastructure unfolds.
