Eviden and AMD Build France’s First Exascale Supercomputer 'Alice Recoque'

Eviden, the Atos Group product brand leading in advanced computing, and AMD announced their selection to build Alice Recoque, a next-generation supercomputer to support scientific computing (HPC) and artificial intelligence (AI), serving as an AI Factory. Alice Recoque will be France’s first and Europe’s second Exascale supercomputer, a machine to expand Europe’s AI and research capabilities while ensuring energy efficiency and sovereignty.

This project, led by France GENCI, operated by CEA, will have the capacity to surpass one Exaflop per second for scientific simulation. This extreme performance, equivalent to more calculations than humanity could accomplish in four years of relentless mental computation, will mark a fiftyfold increase in computing capacity compared to the previous system, while only multiplying by 5 the electrical power.

The project, representing an overall cost of 554 million euros over 5 years of operation, is funded by EuroHPC JU, with budget stemming from the Digital Europe Programme (DEP), and by the Jules Verne consortium, led by France through GENCI and CEA with participation from the Netherlands with SURF and Greece with GRNET.

Alice Recoque will be installed within France’s CEA’s Very Large Computing Center (TGCC), which already hosts Eviden-built systems such as GENCI’s Joliot-Curie machine and CEA’s Topaze machine.

Alice Recoque will tackle Europe’s most pressing societal, scientific, and industrial challenges by combining traditional High-Performance Computing (HPC) and AI workloads from large-scale simulations, data analysis, and AI models. It covers the entire computing lifecycle, integrating cutting-edge hardware and advanced AI software to deliver scalable high-impact solutions. This project is a live implementation of the strong collaboration and commitment from Eviden and AMD to accelerate research and industrialization of HPC-AI use cases, with major investments in human and technological resources.

The acceleration partition of the Alice Recoque system will be powered by next-gen AMD EPYC CPUs, codenamed “Venice,” AMD Instinct MI430X GPUs, and AMD FPGAs. In addition, Alice Recoque will welcome a scalar compute partition based on the European Rhea2 processor developed by SiPearl. The entire system will be interconnected by Eviden’s network solution (BXIv3) into its newest BullSequana XH3500 platform, along with storage solutions provided by DDN.

This ecosystem will enhance climate modeling, accelerate innovation in materials and energy, enable digital twins for personalized medicine, and support next-gen European foundational AI models. It will also address the vast data generated by scientific instruments such as telescopes, satellites, IoT devices, and AI applications, driving breakthroughs across multiple domains.

Alice Recoque will integrate various European technologies to ensure technological sovereignty. The system will use Eviden’s BXIv3 networking technology and an additional European SiPearl Rhea2 CPU partition to increase computation capacities. Nearly three-quarters of BullSequana XH3500 components will be produced in Europe, ensuring traceability, regulatory compliance, and reduced geopolitical risks while aligning with Europe’s climate goals.

Composed of 94 racks, Alice Recoque is expected to be one of Europe’s top supercomputers for double-precision HPC workloads. It will offer exceptional memory performance, enabling deeper insights, faster simulations, and more scientific breakthroughs. With 25% fewer racks and components than other Exascale systems and up to 50% better energy efficiency per GPU, it delivers maximum performance at minimum cost and power, supporting Europe’s green computing goals.

Powered by AMD Instinct MI430X GPUs, Alice Recoque supports advanced AI data types, including FP4 and FP8. Each GPU integrates 432 GB of HBM4 memory and 19.6 TB/s bandwidth, delivering leadership capacity and throughput per GPU. The system is connected via Eviden’s BXIv3 interconnect to optimize distributed communication and resource utilization.

Eviden’s integrated hardware and smart software provide leading computing power with improved application workload energy efficiency, expected to reach about 20% compared to equivalent exascale systems. Real-time monitoring and energy optimization are enabled by Eviden’s Argos software, while 5th generation Direct Liquid Cooling technology uses warm water to cool 100% of rack components.