Arrhenius technical description

Arrhenius is a mid-range EuroHPC supercomputer. It is an HPE Cray EX supercomputer consisting of several partitions, each targeted for different use cases.

The largest partition is Arrhenius GPU which consists of GPU-accelerated compute nodes running Nvidia Grace Hopper Superchips. There is also a smaller CPU-only partition, Arrhenius CPU, featuring AMD EPYC “Turin” CPUs.

The CPU and GPU partitions share a Lustre file system divided into Disk (25 PB, 260 GiB/s) and Flash (2 PB, 15M IOPs for random 4K reads) tiers. Each CPU and GPU compute node has local flash storage that can be used as a node-local scratch file system during a job.

Arrhenius SENS is a dedicated HPC partition for processing sensitive personal data. It is designed to meet strict security requirements for large-scale data storage and handling, in full compliance with European laws and regulations, without significant compromises to usability or performance.

Security includes ISO 27000 series-based measures such as isolated user environments, strong authentication, restricted Internet access, and enhanced activity logging. The partition provides a typical HPC environment with SLURM job management and supports both terminal and graphical logins. Note that transferring sensitive data to remote systems requires a data processing agreement or equivalent arrangement.

Arrhenius PCD (Persistent Compute and Data) is a partition dedicated to persistent services with cloud access. It contains a CPU compute module, a small three-node GPU module with NVIDIA L40S 48GB GPUs for specialised GPU workloads, and a CEPH storage module offering 2.5 PB of raw storage. Collectively, these resources provide over 4,000 CPU cores connected by high-speed Ethernet. The partition supplies persistent compute resources as virtual machines via a virtualisation platform, with plans to add container-based services.

Arrhenius CPU partition

Thin nodes, 192 nodes, each with:

  • CPU: 2x AMD EPYC 9755 Turin 2.7 GHz (128 cores)
  • RAM: 768 GB DDR5
  • Interconnect: 1x Slingshot 200 GB
  • Local Storage: 1.8 TB NVME

Fat nodes, 20 nodes, each with:

  • CPU: 2x AMD EPYC Turin 9745 2.4 GHz (128 cores)
  • RAM: 3 TB DDR5
  • Interconnect: 1x Slingshot 200 GB
  • Local Storage: 1.8 TB NVME

Arrhenius GPU partition

GPU nodes, 382 nodes, each with:

  • CPU: 4x NVIDIA GH200 superchips 72 core ARM
  • GPU: 4x NVIDIA GH200 96 GB HBM 128 GB LPDDR Modules
  • Interconnect: 4x Slingshot 200 GB
  • Local Storage: 1.8 TB NVME

Distributed storage (for CPU and GPU)

Arrhenius CPU and GPU share a single Lustre filesystem. The filesystem is partitioned into a Disk and a Flash part. The Disk part has a capacity of 25 PB and an aggregated performance of 260 GiB/s. The Flash part has a capacity of 2 PB and an aggregated performance of 15M IOPS (random 4K read).

Arrhenius SENS partition

3 Small CPU nodes, each with:

  • 2x 128-core AMD EPYC 9745 at 2.4 GHz and 768 GB RAM.

17 Large CPU nodes, each with:

  • 2x 128-core AMD EPYC 9745 at 2.4 GHz and 3072 GB RAM.

20 Small GPU nodes, each with:

  • 4x Nvidia L40S 48GB GPUs and 2x 32-core AMD EPYC 9335 at 3.0 GHz and 384 GB RAM.

1 Large GPU node, with:

  • 4x Nvidia H100 80GB and 2x 32-core AMD EPYC 9335 at 3.0 GHz and 768 GB RAM.

VAST Storage 10 PB

Arrhenius PCD partition

  • 32 nodes with 2x AMD EPYC 9555, 3.2 GHz, 64 core
  • Three of the above nodes are also equipped with one Nvidia L40S 48 GB GPU
  • 1 CEPH Storage module with 2.5 PB raw storage