HoreKa Ruby
KIT’s high-performance computing system HoreKa is being expanded with HoreKa Ruby, a powerful new partition designed for AI and machine learning applications. The new partition is powered by NVIDIA H200 Tensor Core GPUs, specifically developed for data-intensive, highly scalable workloads and compute-heavy AI and machine learning tasks.
The expansion adds 13 nodes, each equipped with four NVIDIA H200 accelerators, offering 141 GB of memory per GPU. HoreKa Ruby nodes are in operation since August 2025.
| Node Type | HoreKa Ruby (Accelerated) |
|---|---|
| No. of nodes | 13 |
| CPUs | AMD EPYC 9454 |
| CPU Sockets per node | 2 |
| CPU Cores / Threads per node | 96 / 192 |
| Main memory | 768 GB |
| Accelerators | 4x NVIDIA HGX H200 |
| Memory per accelerator | 141 GB |
| Local disks | 15,36 TB NVMe SSD |
| Interconnect | 4x NDR400 |
HoreKa Ruby complements the existing HoreKa Blue, Green and Teal components and integrates seamlessly into the current infrastructure. Users benefit from cutting-edge hardware, high-speed InfiniBand connectivity, and energy-efficient cooling.
This expansion reinforces KIT’s role in national high-performance computing and is accessible to researchers across Germany through NHR∂KIT.
More information about the Artificial Intelligence Partition "HoreKa Ruby" can be found in the user documentation.