Europe's Jupiter Supercomputer Receives Domestic Silicon Boost
SiPearl's Rhea1 CPUs mark a milestone in EU's sovereign computing ambitions
1European silicon milestone
SiPearl has begun delivering its Rhea1 CPUs for integration into Jupiter, Europe's most powerful supercomputer. This marks the first major deployment of European-designed processors in an exascale system. The 80-core Rhea1 chips feature 61 billion transistors and were specifically optimized for HPC workloads. Unlike Jupiter's existing Nvidia-powered booster section, this CPU-only partition represents Europe's sovereign computing capabilities.
2Jupiter's hybrid architecture
The supercomputer currently combines two distinct computing architectures. The record-setting Booster section uses 24,000 Nvidia GH200 superchips with Grace CPUs and H200 GPUs. The new CPU partition will eventually comprise 1,300 nodes with 2,600 Rhea1 processors. While the GPU section delivers extreme performance, the CPU partition handles workloads less suited to parallel processing. This hybrid approach mirrors Japan's Fugaku supercomputer strategy.
3Technical specifications
Each Rhea1 processor contains four stacks of HBM2e memory totaling 64GB with 1.8TB/s bandwidth, plus four DDR5 channels. The Arm Neoverse V1-based design emphasizes memory bandwidth over raw FLOPS, similar to Fugaku's A64FX architecture. This makes it particularly effective for scientific applications requiring efficient data access rather than pure computational throughput. The entire CPU partition is estimated to deliver 5 petaFLOPS.
4Strategic importance
The deployment represents Europe's push for technological sovereignty in high-performance computing. EuroHPC JU highlights this as proof of Europe's ability to combine research, industry and infrastructure. While Rhea1 uses older Arm cores compared to Nvidia's Grace CPUs, its development demonstrates Europe's growing HPC ecosystem. Future systems like Alice Recoque will build on this foundation with next-gen Rhea2 chips.
5Performance considerations
Rhea1's architecture reflects lessons from Fugaku's success in memory-intensive benchmarks. The Top500's HPCG test, which favors balanced systems, historically rewarded Fugaku despite its lower HPL ranking. Jupiter's CPU partition follows this design philosophy, prioritizing memory bandwidth (1.8TB/s per chip) for data-heavy scientific workloads. This complements the GPU section's raw compute power for different application types.
6Industry implications
SiPearl's design win extends beyond Jupiter to the upcoming Alice Recoque system, which will combine Rhea2 CPUs with AMD accelerators. This establishes a pattern of European HPC systems blending domestic CPU designs with leading GPU technologies. The approach balances sovereignty concerns with performance requirements, avoiding complete dependence on any single foreign technology provider.
7Computational balance
Jupiter's architecture acknowledges that not all supercomputing workloads benefit equally from GPU acceleration. While the Nvidia-powered Booster section delivers extreme performance for parallelizable tasks, many scientific applications require the CPU partition's different strengths. This includes workloads with complex dependencies, irregular memory access patterns, or algorithms that don't map well to GPU architectures.
8European HPC roadmap
The Rhea1 deployment represents just the beginning of Europe's sovereign computing initiative. Future systems will incorporate more advanced domestic designs while continuing to leverage best-in-class international components where necessary. This balanced approach aims to build European capabilities without isolating from global technological progress, positioning EU research institutions for both independence and competitiveness.