Introducing Perun¶
Perun, in Slovak mythology the father of Devana, is the new HPC cluster procured by the Computing Centre SAS and put into operation in Q2 of 2026.
The system consists of 45 universal compute nodes totaling 14,400 CPU cores, 15 cloud nodes providing 4,800 CPU cores, and 76 accelerated nodes, each equipped with 4 accelerated CPU+GPU chips, offering a total of 155 TB of RAM. The theoretical peak performance of the system is approximately 17 PFLOP/s, divided into 900 TFLOP/s, 300 TFLOP/s, and 15,200 TFLOP/s for the universal, cloud, and accelerated partitions, respectively.
All universal and cloud compute nodes are built on the BullSequana XH3000 architecture and implemented as BullSequana XH3420 blade servers. Each blade hosts three independent dual-socket compute nodes based on AMD EPYC 5th-generation (Turin) processors using the SP5 platform. High-speed node-to-node communication is provided via InfiniBand NDR. The accelerated partition is built on the 4 NVIDIA Grace Hopper GH200 Superchip architecture. Detailed hardware specifications are described in the Compute Nodes section.
The cluster runs the Rocky Linux 9 operating system (Red Hat Linux family). A wide range of utilities, scientific applications, and programming libraries maintained by the HPC team is available to users. The software stack is generated using EasyBuild and provided via environment modules. For details, see the Software, Compilation, and Environment sections.
Users have access to shared filesystems mounted on the compute
nodes. Persistent user and project data are available under
/home and /projects, respectively, while temporary
high-performance storage for running jobs is provided via
/scratch. Data stored on the /projects mount point are
shared across the Perun and Devana clusters. Additional
information is available in the
Mountpoints section.
Resource allocation and workload execution are managed by the open-source Slurm workload manager. Slurm provides fine-grained control over resource usage and job scheduling. Usage instructions are documented in the Job Submission section and Perun-specific Partitions section.
