Named Gefion and built by Eviden, a subsidiary of the global Atos group, it will be based on Nvidia's DGX SuperPod architecture. Gefion will contain more than 1,500 of Nvidia's H100 Tensor Core GPUs and will deliver six exaflops of FP8 AI performance. Furthermore, it will be connected With the open source software platform Nvidia Cuda Quantum, which allows simulations to be performed on hybrid quantum computers.
The supercomputer will belong to the Danish Center for Artificial Intelligence Innovation and will be hosted by the Digital Realty data center, which uses 100% renewable energy.
Novo Nordisk is investing about 80 million euros in the project through the Novo Nordisk Foundation, and the Danish Export and Investment Fund is also contributing about 8 million euros.
“Scientific discoveries depend on data, and artificial intelligence has presented us with an unprecedented opportunity to accelerate research in, for example, human and planetary health,” Mads Krogsgaard Thomsen, CEO of the Novo Nordisk Foundation, said in a statement.
Explaining the partnership agreement at a press conference, Kimberly Powell, Vice President of Healthcare at Nvidia, said: “In our cooperation agreement, we will take all of this generative AI and move it to the sovereign AI infrastructure so that Denmark can drive the development of medicine and computing.” Quantitative and social sciences.
Meanwhile, AI and the supercomputers that support its training are no longer the domain of a few select researchers, and as the technology becomes more ubiquitous, so does the need for computing to train larger and larger AI models or run sophisticated simulations.
Its importance ranges from productivity and efficiency gains, to military and cybersecurity applications – something that is not going unnoticed by national security policymakers, as evidenced by US export restrictions on devices used to train artificial intelligence to China.
“In the current geopolitical climate, it is important that we strengthen our strategic positions,” said Morten Budskov, Danish Minister of Industry, Business and Finance (although he avoids today’s catchphrase of “digital sovereignty”).
But staying on top of the HPC and AI training game is a tough task, and with the Vivion supercomputer scheduled to be fully operational in 2025, and featuring an Nvidia H100 GPU, it's a testament to how quickly the Vivion supercomputer is moving. Things in the world of HPC and AI.
Nvidia, the third most valuable company in the world, launched its latest artificial intelligence chip last Tuesday, the Blackwell, which it says is 4 times faster than the H100. It will also be integrated into a superior chip with two Blackwell graphics processing units and a Grace CPU. For “Generative AI for Trillion-Scale Parameters.”