University of Stuttgart and HPE to build exascale supercomputer

• 20 Dec 2023
• TM Team
• 0 Comments

The University of Stuttgart and Hewlett Packard Enterprise (HPE) have announced an agreement to build two new supercomputers at the High-Performance Computing Center of the University of Stuttgart (HLRS).

In the first stage, a transitional supercomputer, called Hunter, will begin operation in 2025. This will be followed in 2027 with the installation of Herder, an exascale system that will provide a significant expansion of Germany’s high-performance computing (HPC) capabilities. Hunter and Herder will offer researchers world-class infrastructure for simulation, artificial intelligence (AI), and high-performance data analytics (HPDA) to power cutting-edge academic and industrial research in computational engineering and the applied sciences.

The total combined cost for Hunter and Herder is €115 million. Funding will be provided through the Gauss Centre for Supercomputing (GCS), the alliance of Germany's three national supercomputing centers. Half of this funding will be provided by the German Federal Ministry of Education and Research (BMBF), and the second half by the State of Baden-Württemberg's Ministry of Science, Research, and Arts.

Hunter to Herder: a two-step climb to exascale

Hunter will replace HLRS’s current flagship supercomputer, Hawk. It is conceived as a stepping stone to enable HLRS’s user community to transition to the massively parallel, GPU-accelerated structure of Herder.

Hunter will be based on the HPE Cray EX4000 supercomputer, which is designed to deliver exascale performance to support large-scale workloads across modeling, simulation, AI, and HPDA. Each of the 136 HPE Cray EX4000 nodes will be equipped with four HPE Slingshot high-performance interconnects. Hunter will also leverage the next generation of Cray ClusterStor, a storage system purpose-engineered to meet the demanding input/output requirements of supercomputers, and the HPE Cray Programming Environment, which offers programmers a comprehensive set of tools for developing, porting, debugging, and tuning applications.

Hunter will raise HLRS’s peak performance to 39 petaFLOPS (39*1015 floating point operations per second), an increase from the 26 petaFLOPS possible with its current supercomputer, Hawk. More importantly, it will transition away from Hawk’s emphasis on CPU processors to make greater use of more energy-efficient GPUs.

Hunter will be based on the AMD Instinct™ MI300A accelerated processing unit (APU), which combines CPU and GPU processors and high-bandwidth memory into a single package. By reducing the physical distance between different types of processors and creating unified memory, the APU enables fast data transfer speeds, impressive HPC performance, easy programmability and great energy efficiency. This will slash the energy required to operate Hunter in comparison to Hawk by approximately 80% at peak performance.