News: 1763467442

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Eviden set to build France's first exascale supercomputer with AMD at the wheel

(2025/11/18)


SC25 France will get its first exascale supercomputer — Europe's second — when Atos subsidiary Eviden builds Alice Recoque using AMD chips.

The EuroHPC Joint Undertaking is funding the €554 million ($641.4 million) project to expand European research capabilities in simulations, data analysis, and AI for societal, scientific, and industrial challenges.

Named after French computer scientist Alice Recoque, the modular system will launch with AMD's upcoming "Venice" Epyc CPUs and Instinct MI430X GPUs.

[1]

A second partition will add a CPU cluster like Europe's first exascale system, [2]Jupiter , which runs Nvidia Grace-Hopper GH200 superchips and will expand with [3]SiPearl's Rhea1 processor .

[4]

[5]

Located at France's National High-Performance Computing Equipment ( [6]GENCI ) facility in Paris, the system comprises 94 racks of Eviden's [7]BullSequana XH3500 architecture - upgraded with fanless direct liquid cooling to handle denser, hotter infrastructure.

The accelerator partition will also feature AMD-made FPGAs and will be linked by Eviden's high-bandwidth BullSequana eXascale Interconnect (BXI) networking, v3 of which supports speeds up to 800 Gbps (in the switches – the adapters are 400 Gbps), plus DDN storage.

[8]Europe joins US as exascale superpower after Jupiter clinches Top500 run

[9]Power: The answer to and source of all your AI datacenter problems

[10]UK waves £750M supercomputer contract at HPC builders

[11]Europe's Jupiter supercomputer hits exascale threshold at inauguration

It's expected by late 2026, pending AMD's chip availability next year.

The general-purpose compute partition, based on SiPearl's Rhea2 CPU and BXI networking, will arrive in 2027, although SiPearl hasn't yet shipped production versions of its first processor.

[12]

Eviden claims Alice Recoque will operate at just 12 megawatts for average workloads, making it more energy-efficient than comparable exascale systems. While Eviden calls it an "AI Factory" covering the full AI lifecycle, this differs from Nvidia's [13]definition . Alice Recoque will also handle traditional HPC workloads like modeling and simulations.

The company says that Alice Recoque has been "designed as an AI Factory," covering the entire AI lifecycle. However, it is clearly using a different definition of the term from GPU giant Nvidia, which regards an "AI Factory" as being [14]entirely dedicated to processing AI , while Alice Recoque will also carry out traditional HPC workloads such as modeling and simulations. ®

Updated at 16.42 UTC to add:

[15]

Eviden has made contact following publication of this article to query its own claims of Alice Recoque operating at 12 megawatts. A spokesperson told us:

"This figure was part of a preliminary version and is not an exact science, rather an expectation that may vary depending on the production capacity."

Get our [16]Tech Resources



[1] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/202511sycompsupercomputing&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aRymJ7n9MRch4ehwlejl1QAAAIk&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[2] https://www.theregister.com/2025/11/17/europe_jupiter_supercomputer/

[3] https://www.theregister.com/2025/07/09/sipearl_rhea1_tape_out/

[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/202511sycompsupercomputing&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aRymJ7n9MRch4ehwlejl1QAAAIk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/202511sycompsupercomputing&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aRymJ7n9MRch4ehwlejl1QAAAIk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[6] https://gnius.esante.gouv.fr/en/players/player-profiles/national-high-performance-computing-equipment-genci

[7] https://eviden.com/solutions/high-performance-computing/hpc-solutions/bullsequana-xh3500-supercomputer/

[8] https://www.theregister.com/2025/11/17/europe_jupiter_supercomputer/

[9] https://www.theregister.com/2025/11/15/power_supercomputing/

[10] https://www.theregister.com/2025/10/13/edinburgh_supercomputer_procurement/

[11] https://www.theregister.com/2025/09/05/jupiter_supercompute_exascale_threshold/

[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/202511sycompsupercomputing&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aRymJ7n9MRch4ehwlejl1QAAAIk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[13] https://www.theregister.com/2025/10/29/nvidia_omniverse_dsx/

[14] https://www.theregister.com/2025/10/29/nvidia_omniverse_dsx/

[15] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_specialfeatures/202511sycompsupercomputing&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aRymJ7n9MRch4ehwlejl1QAAAIk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[16] https://whitepapers.theregister.com/



That means I'll have to use $ans to suppress newlines now.
Life is ridiculous.
-- Larry Wall in Configure from the perl distribution