Sweden to add more Nvidia systems to Berzelius supercomputer, turning it into one of world’s most powerful AI systems
Sweden is to upgrade its Berzelius supercomputer by adding more high-end Nvidia systems, the graphics chipmaker said.
The supercomputer is currently based on 60 Nvidia DGX A100 systems, and Sweden plans to add another 34, Nvidia said.
The DGX A100 systems include graphics processing units (GPUs) with 80GB of memory each.
The upgraded system is planned to deliver nearly half an exaflop of AI performance, making it one of the world’s 100 fastest AI supercomputers, Nvidia said in a blog post.
AI language models
Berzelius is based on Nvidia’s DGX SuperPOD AI data centre infrastructure platform.
The system is operated by the Wallenberg Artificial Intelligence, Autonomous Systems and Software Programme (WASP), Sweden’s biggest research effort, which includes five partner universities and other affiliated research groups.
The majority of WASP’s funding is provided by the Knut and Alice Wallenberg Foundation (KAW), a public and private foundation formed in 1917 to support natural sciences, technology and medical research.
Last June the Berzelius system – named after groundbreaking chemist Jacob Berzelius – was used to train GPT-SW3, a family of large language models capable of drafting a speech or answering questions in Swedish.
Researchers plan in the longer term to train a version of the model with 173 billion parameters that is fluent in languages such as Danish and Norwegian.
A data-driven life sciences programme funded by the Wallenberg Foundation intends to use Berzelius to understand diseases such as cancer, improve the detection of cancer cells and navigate the data found in human genomes, Nvidia said.
Amidst a global economic slowdown artificial intelligence is one area that is continuing to attract funding, with Microsoft on Monday announcing a multi-billion dollar, multi-year deal with OpenAI.