JUPITER Supercomputer Just Simulated a Brain in 5 Days—And It's Creepy
Europe's JUPITER supercomputer just crushed a brain-mapping task that would take a mortal eternity, using Nvidia silicon to turn biological data into CytoNet. It’s a brilliant leap for neuroscience, or perhaps just a very expensive digital ghost experiment.
JUPITER, Europe's first exascale supercomputer, successfully trained CytoNet, a foundational model of brain microarchitecture, in just 120 hours. The project, led by Katrin Amunts and Christian Schiffer at the Jülich Research Centre, ingested 6.5 petabytes of data harvested from 21 deceased brains using 4,096 Nvidia Grace Hopper superchips.
The model maps the brain at a cellular level, linking microscopic structures to macro-level functions. With 86 billion neurons and 100 trillion synaptic connections to account for, the system acts as an autonomous agent that generates its own scientific hypotheses rather than just crunching numbers. The team plans to evolve this into an AI assistant for neuroscientists, utilizing models like Nvidia Nemotron 3 120B to allow researchers to query brain data via conversational interfaces.
The same hardware recently set a world record in climate modeling with ICON, simulating 146 days of Earth’s climate in just one day, and successfully emulated a 50-qubit quantum machine (JUQCS-50). It turns out the best way to prove quantum computing works is to have a classical supercomputer pretend to be one, highlighting the slightly ironic reality that our most advanced 'quantum' breakthroughs are currently running on traditional Nvidia-powered silicon.
Source: Nvidia Blog
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.