JUPITER exascale powers brain mapping, climate, 6G and quantum

Forschungszentrum Jülich and partners announce results from the JUPITER exascale supercomputer demonstrating production-scale science across neuroscience, climate, communications and quantum simulation.

• Main announcement: JUPITER (Forschungszentrum Jülich) has run four major projects at exascale: the CytoNet foundation model for brain microarchitecture, a 1-km ICON coupled Earth-system climate simulation, a research collaboration with Ericsson for 5G/6G AI, and JUQCS-50 — a full 50-qubit quantum computer simulation. Key facts: CytoNet trained in under five days on 4,096 NVIDIA Grace Hopper Superchips using 6.5 PB of data from 21 post-mortem brains; ICON ran on 20,480 Grace Hopper Superchips and simulated ~146 days of real climate in 24 hours; JUQCS-50 surpassed the previous 48-qubit simulation record.
• Background & partners: The brain work is led by INM-1 (Katrin Amunts, Christian Schiffer) with Helmholtz AI and Helmholtz partners; the ICON configuration was developed with ETH Zurich, DKRZ, JSC, Max Planck Institute for Meteorology, NVIDIA, CSCS, and University of Hamburg and won the Gordon Bell Prize (SC25). The Ericsson collaboration was announced in March and focuses on energy-efficient, brain-inspired AI for radio/core networks; JUQCS-50 is available via the JUNIQ quantum user facility at JSC.