Datacenter Forum
updated: 11-11-2020 | 15:19
The Knut and Alice Wallenberg Foundation has donated SEK 300 million to Linköping University and the National Supercomputer Centre (NSC) to build Sweden´s fastest supercomputer for machine learning and artificial intelligence.
The construction and installation of the BerzeLiUs supercomputer, named after scientist Jacob Berzelius from Östergötland, will begin at the start of 2021. The supercomputer, a NVIDIA DGX SuperPOD, will offer a processing speed of 300 petaflops for AI.
“I am extremely happy and proud that Linköping University will, through the National Supercomputer Centre, be host for this infrastructure”, says Jan-Ingvar Jönsson, vice-chancellor of Linköping University. “This gives us confidence that Sweden is not simply maintaining its international position, but also strengthening it.”
AI computing
Contracts have been signed between NSC and Atos, a global company within digital transformation, which is responsible for delivery and installation. BerzeLiUs will initially be equipped with 60 of the latest and most rapid servers, DGX systems, with eight graphics processor units (GPUs) in each. These will be delivered by NVIDIA - a global company in AI computing. The servers will be combined with networks from NVIDIA, an application toolset from Atos, and 1.5 petabytes of data storage (see more details below).
BerzeLiUs will be a powerful resource to advance AI research and boost collaboration between academia and Swedish industry. This will primarily be within research programmes financed by the Knut and Alice Wallenberg Foundation, such as the Wallenberg Artificial Intelligence, Autonomous Systems and Software Program and initiatives in the life sciences and quantum technology.
“The new supercomputer will bring AI research in Sweden into a whole new era. It will be the centre for an investment that not only raises Sweden’s position as leader in AI research but also gives Swedish industry great competitive advantages in areas such as telecommunication, pharmaceuticals and vehicles”, says Jaap Zuiderveld, vice president EMEA at Nvidia.
Turnkey installation
It has previously taken months or even years of planning and construction to commission this type of AI infrastructure. An NVIDIA DGX SuperPOD is an integrated turnkey solution that can be delivered and installed in a few weeks.
“We are proud to be working with Linkoping University on the delivery of their new supercomputer BerzeLiUs, which will provide researchers with the computing power to enable new scientific breakthroughs and innovation. This high-performance system, coupled with our expertise, will help Sweden to address key AI and machine learning challenges”, says Damien Déclat, business area manager for high-performance supercomputers at Atos.
The National Supercomputer Centre at LiU is today home to six traditional supercomputers with a total processing speed of around 6 petaflops. The largest of these, Tetralith, was the most powerful supercomputer in the Nordic region when it was installed in the summer of 2018. The processing speed required for the algorithms that are used in AI and machine learning cannot, however, be compared with the capacity of traditional supercomputers:
“BerzeLiUs will be more than twice as fast as Tetralith. This is a truly powerful AI resource and by far the most rapid processing cluster we have installed”, Niclas Andersson, Technical Director at the National Supercomputer Centre, confirms.
Technical specifications
BerzeLiUs, or the NVIDIA DGX SuperPOD™, will contain 60 NVIDIA DGX™ A100 systems combined with NVIDIA Mellanox™ InfiniBand networking and 1.5 petabytes of data storage from the company DDN. The Atos toolset Codex AI Suite will support researchers in using the system efficiently.
