Overview
- After months of trials, the booster module has entered formal operation, with a June Top500 submission of 793 petaflops sustained and an FP64 target of about 1.4 exaflops once early issues are resolved.
- The machine integrates roughly 24,000 Nvidia GH200 modules built by ParTec‑Eviden, underscoring reliance on U.S. chip designs while a smaller Rhea1‑based cluster is planned for 2026–2027.
- Positioned for climate, medical and materials research, the system will also serve as the core of the Jupiter AI Factory, offering vastly higher mixed‑precision throughput for training large AI models.
- Project outlays over the coming years total about €500 million, with dense warm‑water cooling and planned campus heat reuse supporting what operators call the most energy‑efficient exascale system among the top five.
- Operators acknowledge early hardware faults and outages since May and say manufacturers are replacing parts under contract while the system continues running; a widely shared WELT item about a shutdown is labeled satire.