Skip to content

Node types

HoreKa 2 is a hybrid system combining separate x86 and ARM-based hardware islands. This design allows different workloads to run on the architecture that best matches their software and performance requirements.

Partition overview

The cluster is divided into several partitions with different node types and accelerator configurations. These partitions are optimized for CPU-only workloads, memory-intensive tasks, and GPU-accelerated research computing.

CPU partitions

  • HoreKa Onyx: standard CPU partition for broadly compatible x86 workloads.
  • HoreKa Blue: memory-optimized CPU partition for large-memory applications.

GPU partitions

  • HoreKa Jade: accelerated partition based on NVIDIA GB200 NVL4 platforms.
  • HoreKa Teal: accelerated partition using NVIDIA H100 and H200 GPU nodes.
  • HoreKa Ruby: accelerated partition using NVIDIA H200 GPU nodes.

Hardware characteristics

The node classes differ in processor architecture, memory capacity, accelerator count, and local storage. These differences are designed to match a broad range of scientific workflows, from tightly-coupled MPI jobs to memory-heavy simulations and AI training workloads.

For the detailed node matrix and partition information, see the main HoreKa 2 overview.