Jupiter Supercomputer Integrates SiPearl Rhea1

Germany's Jupiter supercomputer advances toward Exascale performance by integrating SiPearl Rhea1 CPUs into its 162-node Universal Cluster.

Germany's Jupiter supercomputer is advancing toward Exascale performance with the integration of SiPearl's Rhea1 CPUs. The Universal Cluster will feature 162 nodes, each equipped with two 80-core Rhea1 processors manufactured on TSMC's N6 process technology.

The Rhea1 chip includes 64GB of HBM2e and DDR5 memory, with official performance data expected in autumn 2026. Jupiter also utilizes a GPU booster with 5,884 nodes, each housing four Nvidia GH200 units, providing a total performance exceeding 1 ExaFLOP.

The system includes 582 nodes with dual 96-core AMD Epyc 9655 CPUs. Looking ahead, the French Alice Recoque Exascale system will debut in 2026 using Rhea2 chips, AMD Epyc 9006 Zen-6 processors, and AMD Instinct MI430X units with 432GB of HBM4 memory.