0
Chancellor Friedrich Merz asked on Friday (5/9) that Europe recovers the delay in the artificial intelligence sector by inaugurating the first “Exascale” supercomputer of the continent, a term used in high performance computing (supercomputation) to designate systems capable of performing at least 1 “Exflop”-a quintillion (10¹⁸) of floating point operations per second.
“And they compete side by side the leadership in a world economy driven by artificial intelligence,” Merz said in Jülich, near Colony, in the center that houses the new supercomputer “Jupiter” (“Jupiter” in the original spelling) “but Germany and Europe have all the conditions to achieve this level.”
Read too
The equipment, considered a technological milestone, should help reduce European lag in the field of AI and boost scientific research, especially on climate change.
What is Jupiter?
Installed in Jülich, west of colony, Jupiter is the first “Exascale” supercomputer in Europe, with the capacity to perform at least one quintal of calculations per second. The United States already has three machines in this category, all operated by the Department of Energy.
With about 3,600 square meters-almost half of a soccer field-the system brings together processor rows and about 24,000 NVIDIA NVIDIA chips, widely used in the AI industry.
Built by the French Acts, at the cost of € 500 million, the project was funded equally by the European Union and the German government. He is the first exascale supercomputer in Europe and the fourth in the world, according to available data.
“We are talking about what is most powerful today on high performance computers,” said Emmanuel Le Roux, Actos Advanced Computing Director.
For Thomas Lippert, director of the Jülich Center, it is a “performance jump” to the continent: “The machine is 20 times more powerful than any other computer in Germany.”
AI race
Jupiter is the first European supercomputer considered competitive for training artificial intelligence models, according to Lippert.
A report from Stanford University released this year points out that in 2024 US institutions created 40 prominent AI models, against 15 from China and only three from Europe.
“The new system has a huge weight in European efforts to train AI models,” said Jose Maria Cela, a researcher at the Barcelona Supercomputation Center. “The performance of a model depends directly on the calculation capacity of the computer used,” he told AFP, adding that Europe suffers from an infrastructure “deficit” in this size.
Jupiter will provide the power needed to train Large Language Models (LLMs), which generate massive text volumes and feed chatbots such as chatgpt and Gemini.
Still, the machine is highly dependent on Nvidia’s American technology at a time of increasing friction between the United States and Europe.
Read more in, partner of the Metropolis.
