{"id":1397362,"date":"2025-11-19T10:06:45","date_gmt":"2025-11-19T10:06:45","guid":{"rendered":"http:\/\/aRkQdnGDTkuT9pyRLgweqF"},"modified":"2025-11-19T10:06:45","modified_gmt":"2025-11-19T10:06:45","slug":"amd-is-the-hardware-of-choice-for-frances-first-ai-factory-supercomputer","status":"publish","type":"post","link":"https:\/\/arcader.org\/news\/amd-is-the-hardware-of-choice-for-frances-first-ai-factory-supercomputer\/","title":{"rendered":"AMD is the hardware of choice for France’s first ‘AI factory’ supercomputer"},"content":{"rendered":"
\n

Move over Nvidia\u2014France’s first Exascale Supercomputer will be powered by AMD hardware. Eyponomously named the Alice Recoque, after the French computer scientist who passed away in 2021, the supercomputer is a collaboration between AMD and Eviden<\/a>.<\/p>\n

Europe’s second supercomputer to date, the Alice Recoque will be built with next-gen AMD EPYC CPUs, codenamed “Venice,” and AMD Instinct MI430X GPUs. The latter of these is part of the MI400 series of accelerators that is geared towards AI computing for scientific research specifically. Eviden then covers networking, connecting AMD’s components via its BullSequana XH3500 platform. The whole thing is being touted as an ‘AI factory’.<\/p>\n

The hope is that all of this cutting-edge tech will mean France’s first supercomputer will be capable of delivering “more than one exaflop of HPL performance” upon completion. HPL here refers to ‘High-Performance Linpack’, which is a type of benchmark to test serious processing power, such as what the Alice Recoque will eventually possess.<\/p>\n

According to AMD’s own definition of HPL<\/a>, this benchmark specifically involves solving “a (random) dense linear system in double precision (64 bits) arithmetic on distributed-memory computers.”<\/p>\n

I won’t attempt to drill down into the specifics on that one, but it’s a benchmark specifically used by supercomputers to deliver a big number on which to hang their absurdly high costs. Sort of like a car’s top speed, it’s unlikely to hit this sorta performance all the time, but it’s undoubtedly extremely powerful.<\/p>\n

Let’s return to a subject a little more approachable: money. The entire project will cost an estimated \u20ac544 million. Funding is a team effort stemming in part from the Digital Europe Programme and the Jules Verne Consortium, which includes partner organisations in France, the Netherlands, and Greece.<\/p>\n

\n
\n
\n

\"Eviden's<\/p>\n<\/div>\n<\/div>

Eviden’s concept drawing for the Alice Recoque. Shame it won’t actually look like this. <\/span>(Image credit: Eviden)<\/span><\/figcaption><\/figure>\n

The hope is that all those exoflops will eventually “enhance climate modeling, accelerate innovation in materials and energy, enable digital twins for personalized medicine, and support next-gen European AI models.”<\/p>\n

Long story short, there’s a lot of money and a lot of international interest wrapped up in this project. This is in the hopes of maintaining “sovereignty” in Europe’s own AI research, thereby avoiding a reliance on, say, AI technologies founded by US companies. However, if Google CEO Sundar Pichai’s recent warning is to be believed\u2014that “No company is going to be immune<\/a>” in the event of an AI bubble burst\u2014it’s hard to predict just how important ‘sovereignty’ in AI research will be.<\/p>\n

\n