NVIDIA has announced the JUPITER class of supercomputers, a groundbreaking AI-driven system for scientific advancements, set to be powered by the NVIDIA Grace Hopper accelerated computing architecture. This architecture is designed to provide unparalleled computing power for AI and simulation workloads.
NVIDIA has announced the JUPITER class of supercomputers, a groundbreaking AI-driven system for scientific advancements, set to be powered by the NVIDIA Grace Hopper accelerated computing architecture. This architecture is designed to provide unparalleled computing power for AI and simulation workloads.
The JUPITER supercomputer, hosted at the Forschungszentrum Jülich in Germany, represents a collaborative effort involving NVIDIA, ParTec, Eviden, and SiPearl. It’s aimed at accelerating the development of foundational AI models across various fields, including climate and weather research, material science, drug discovery, industrial engineering, and quantum computing. JUPITER marks the debut of a quad NVIDIA GH200 Grace Hopper Superchip node configuration.
Key features of JUPITER include:
- Quad NVIDIA GH200 Grace Hopper Superchip Node Configuration: This configuration is based on Eviden’s BullSequana XH3000 liquid-cooled architecture. It includes a booster module with nearly 24,000 NVIDIA GH200 Superchips, interconnected with NVIDIA’s Quantum-2 InfiniBand networking platform. This setup positions JUPITER as the world’s most powerful AI system, capable of delivering over 90 exaflops of AI training performance — a staggering 45 times more than the previous Jülich’s JUWELS Booster system. It also achieves one exaflop for high-performance computing (HPC) applications, all while consuming just 18.2 megawatts of power.
- Innovative Node Architecture: The quad GH200 features 288 Arm Neoverse cores and can achieve 16 petaflops of AI performance using up to 2.3 terabytes of high-speed memory. This is facilitated by four GH200 processors networked through a high-speed NVIDIA NVLink® connection.
Ian Buck, NVIDIA’s Vice President of Hyperscale and HPC, highlighted that the JUPITER supercomputer, leveraging NVIDIA GH200 and advanced AI software, will deliver exascale AI and HPC performance. This development is expected to significantly advance scientific and technological frontiers.
JUPITER’s capabilities extend to various critical areas:
- Climate and Weather Prediction: Utilizing NVIDIA Earth-2, JUPITER will enhance high-resolution climate and weather simulations, contributing to global initiatives like the Earth Virtualization Engines (EVE).
- Drug Discovery: The NVIDIA BioNeMo™ and NVIDIA Clara™ platforms will streamline the development and deployment of models essential to drug discovery.
- Quantum Computing Technologies: NVIDIA’s cuQuantum software development kit and CUDA® Quantum platform will facilitate significant advancements in quantum computing research.
- Industrial Engineering: The NVIDIA Modulus framework and NVIDIA Omniverse™ platform will revolutionize engineering design, development, and manufacturing processes through AI-accelerated simulation and digital twins.
The installation of the JUPITER system is scheduled for 2024, marking a significant milestone in the field of AI supercomputing and its application in solving some of the world’s most challenging problems.
Engage with StorageReview
Newsletter | YouTube | Podcast iTunes/Spotify | Instagram | Twitter | TikTok | RSS Feed
