Humboldt-Universität zu Berlin - Artificial Intelligence at HU Berlin

High Performance Computing Cluster (HPC Cluster)

With HPC@HU, the CMS of the HU provides an innovative high-performance computing service to make hardware and software for high-performance computing easily accessible to all researchers and teachers at the HU in a centralized and resource-efficient manner. The HU is thus creating a flexible and directly usable Tier 3 HPC offering that meets the very different requirements of the HU's many disciplines. HPC@HU is tackling this challenge with a virtualized infrastructure based on the open source software OpenStack.

The Azimuth interface facilitates the creation of platforms such as:

- Jupyter-HUB,

- Jupyter notebooks,

- Kubernetes clusters,


- R-Studio or

- Linux VMs.


The HPC cluster currently consists of 35 nodes with a total of 50 TB RAM and 4480 CPU cores. Of these, 30 nodes are equipped with GPUs and offer 2.3 TB of GPU memory as well as 12,960 Tensor cores and 207,460 CUDA cores in total. A shared scratch memory of 1.5 PB is available via a Lustre installation and 15 TB locally per node.