SourceStalecollected in 5h

JUPITER Supercomputer Demonstrates Exascale Scientific Capabilities

Read original on NVIDIA Blog
#exascale#hpc#supercomputing

See how exascale computing and Grace Hopper chips are solving the world's most complex scientific challenges.

30-Second TL;DR

What Changed

JUPITER is Europe's first exascale supercomputer located at Forschungszentrum Jülich.

Why It Matters

This deployment marks a milestone for European research, providing the necessary compute for large-scale AI and simulation models.

What To Do Next

Monitor the Forschungszentrum Jülich project updates to understand how exascale hardware accelerates large-scale scientific AI workflows.

Who should care:Researchers & Academics

Key Points

  • •JUPITER is Europe's first exascale supercomputer located at Forschungszentrum Jülich.
  • •The system utilizes NVIDIA Grace Hopper Superchips for high-performance computing.
  • •Infrastructure includes NVIDIA Quantum-X800 InfiniBand networking for massive data throughput.

Deep Insight

AI-generated analysis for this event — not the original article.

Enhanced Key Takeaways

  • •JUPITER is built on the modular supercomputing architecture developed by the DEEP projects, allowing it to combine a general-purpose booster module with specialized accelerator modules.
  • •The system is hosted by the Jülich Supercomputing Centre (JSC) and is a flagship project of the EuroHPC Joint Undertaking, a public-private partnership involving the EU and participating countries.
  • •JUPITER is designed to achieve an HPL-AI benchmark performance exceeding 1 exaflop, specifically targeting AI-driven scientific discovery alongside traditional simulation.
  • •The supercomputer utilizes a liquid-cooling system to manage the high thermal density generated by the massive deployment of NVIDIA Grace Hopper Superchips.
  • •The project integrates the 'JUPITER Exascale Ready' program, which allows researchers to optimize their codebases for the system's unique architecture prior to full-scale deployment.

Competitor Analysis

Architecture
JUPITER (EuroHPC)
NVIDIA Grace Hopper
Frontier (ORNL)
AMD Instinct MI250X
Aurora (ANL)
Intel Max Series GPU
Peak Performance
JUPITER (EuroHPC)
1 Exaflop (FP64)
Frontier (ORNL)
1.2 Exaflops (FP64)
Aurora (ANL)
1 Exaflop (FP64)
Primary Focus
JUPITER (EuroHPC)
AI & Simulation
Frontier (ORNL)
Scientific Simulation
Aurora (ANL)
AI & Simulation

Technical Deep Dive

  • Architecture: Modular Supercomputing Architecture (MSA) consisting of a Booster Module and a Cluster Module.
  • Compute Nodes: Equipped with NVIDIA GH200 Grace Hopper Superchips, combining an Arm-based Grace CPU and Hopper GPU via NVLink-C2C.
  • Interconnect: NVIDIA Quantum-X800 InfiniBand platform providing 800Gb/s throughput per port.
  • Cooling: Direct-to-chip liquid cooling technology to handle high-density power requirements.
  • Software Stack: Integration with NVIDIA AI Enterprise and specialized EuroHPC software environments for exascale scalability.

Future ImplicationsAI analysis grounded in cited sources

JUPITER will accelerate European sovereignty in AI research.
By providing domestic exascale infrastructure, European researchers reduce reliance on non-EU cloud and compute resources for sensitive scientific data.
The system will set a new standard for energy-efficient exascale computing.
The combination of Grace Hopper's power efficiency and advanced liquid cooling is expected to yield a superior performance-per-watt ratio compared to previous generation systems.

Timeline

2022-06
EuroHPC JU announces the selection of Forschungszentrum Jülich as the site for Europe's first exascale supercomputer.
2023-10
NVIDIA and Eviden are selected to build the JUPITER supercomputer.
2024-05
Initial installation phases of the JUPITER infrastructure begin at the Jülich Supercomputing Centre.
2026-06
JUPITER demonstrates exascale scientific capabilities at the ISC High Performance conference.

Weekly AI Recap

Read this week's curated digest of top AI events →

AI-curated news aggregator. All content rights belong to original publishers.
Original source: NVIDIA Blog ↗

This is a summary, not the original. Read the source, or get the weekly briefing.

The weekly digest

One email a week. Unsubscribe anytime.