The modernization was carried out in three partitions, each adapted to their respective workloads. The MPP subcluster, which is currently entering routine operation, consists of 37 compute nodes divided into two partitions. 27 nodes operate in standard "shared" mode, capable of handling multiple jobs simultaneously. Ten nodes run in "single" mode; they are intended to be used for particularly highly-parallelized and optimized workloads that benefit from adaptations to the hardware architecture.
Already in September of last year, nine additional SMP nodes ("Shared Memory Processing") and four GPU nodes ("Graphics Processing Unit") were deployed. The SMP machines each have 1 TB of RAM and are suitable for particularly memory-intensive applications. The GPU nodes each serve as a base for four NVIDIA-H200 high-performance graphics processors (4 x 140 GB of graphics memory), offering powerful acceleration for modern workloads like machine learning. More information about the partitions and their optimal use can be found on the HPC Cluster's documentation page [1].
We would like to express our sincere gratitude to everyone involved in the successful DFG application, planning, and implementation of the new infrastructure for their valuable and dedicated collaboration. We are very pleased with this significant expansion of the cluster resources.
The central computing cluster is being used by more than 1,000 researchers at LUH.
Access requirements and documentation for the "Scientific Computing" service can be found here: [2], [3]. For questions about the computing cluster, please contact cluster-help@luis.uni-hannover.de.
More and more institutes are taking advantage of the opportunity to procure their own hardware and integrate it into the cluster following the conditions of the Research Cluster Housing (FCH) service [4].