Infrastructure¶

HoreKa 2 is housed in a dedicated computer building on KIT's North Campus, built for the high power density and thermal loads of modern HPC systems. The facility is designed to support dense CPU and GPU compute environments while maintaining stable operating conditions, efficient cooling, and strong energy efficiency.

Computer building¶

The cluster is operated in a purpose-built data center environment specifically designed for high-performance computing. This facility combines robust infrastructure for power delivery, ventilation, and thermal management with the flexibility required for large-scale GPU and CPU workloads. The building layout is optimized for modern HPC hardware, including high-density racks and large thermal loads.

Cooling concept¶

HoreKa 2 uses a direct liquid cooling (DLC) concept based on hot-water cooling. Heat generated by the compute nodes is removed directly from the components using liquid cooling loops, with the heat then transferred away in the form of warm water. This approach is especially well suited for dense CPU and GPU systems, where conventional air cooling would become increasingly inefficient at the required power densities.

The hot-water cooling concept improves thermal efficiency and helps maintain stable operating conditions even under sustained high-performance workloads. It is a key element of the overall system design and supports reliable operation of the cluster while reducing the burden on conventional room-cooling systems.

Energy efficiency¶

The facility design and the use of efficient cooling technology are an important part of HoreKa 2's energy performance. The system is specifically engineered to achieve excellent power usage effectiveness (PUE) and water usage effectiveness (WUE), reflecting both the efficiency of the cooling architecture and the overall energy-conscious operation of the facility.

These performance characteristics are particularly relevant for a system such as HoreKa 2, where large CPU and GPU installations consume substantial power and produce considerable heat. The combination of high-density compute hardware and a hot-water DLC cooling concept supports strong efficiency metrics while preserving the performance required for modern research workloads.

Sustainability¶

The infrastructure is designed not only for high performance, but also for sustainable operation. Efficient thermal management, reduced dependence on energy-intensive air-cooling, and optimized facility design contribute to lower operational overhead and improved overall resource efficiency. This is an important aspect of the cluster’s long-term operation in a research and academic environment.