NAISS System News

Arrhenius

The procurement process for Arrhenius, the new EuroHPC system for Sweden, was completed early this summer. It will be an HPE system quite similar to the Dardel system at PDC and will have the same NVIDIA Grace Hopper GPU nodes as in the Dardel expansion (which is described below) plus AMD CPUs that are more recent and powerful than those in Dardel.

Arrhenius will have a CPU partition with 424 AMD Turin 128-core CPUs and a GPU partition with 382 Grace Hopper nodes. The GPU nodes will each have four NVIDIA GH200 Grace Hopper Superchips, and each of those superchips will have a Hopper GPU that is tightly connected with a Grace ARM CPU. The GPU partition is expected to have a High-Performance Linpack (HPL) benchmark performance of more than 60 PFLOPS. Storage will be provided by a 29PB fast parallel Lustre file system. Arrhenius will also have a partition for cloud computing and a partition for sensitive data, each with its own storage.

The system is currently being installed in Linköping and will be completed in the spring of 2026. Arrhenius is planned to replace existing NAISS systems, such as Tetralith (in mid-2026) and Dardel (at the end of 2026).

Dardel

In the previous edition of the PDC Newsletter, we announced that the Dardel system would get more disk capacity plus some more nodes with NVIDIA GPUs. Many researchers requested NVIDIA GPU nodes so they could use applications that are only available for the NVIDIA software stack with CUDA. These upgrades have now been installed, and the new storage and nodes are operational. The Lustre file system has been expanded with an additional 4.7 PB of hard disk storage and 260 TB flash storage. This will give a total usable Lustre capacity of around 22 PB.

Eight Grace Hopper nodes have been added to the system, resulting in a total of 32 superchips. One of the nodes is reserved as a login node.