Germany’s High-Performance Computing Center (HLRS) at the University of Stuttgart has unveiled its latest supercomputer, Hunter. The €15 million system, funded by Baden-Württemberg’s Ministry for Science, Research, and Art alongside Germany’s Federal Ministry of Education and Research, will support a wide array of research endeavors, from engineering and weather modeling to biomedical studies, materials science, and AI.

Constructed by Hewlett Packard Enterprise (HPE), Hunter is built on the Cray EX4000 platform and leverages AMD’s Instinct MI300A accelerated processing units (APUs) paired with Epyc Genoa CPUs. While its architecture mirrors the El Capitan supercomputer’s design, Hunter is significantly smaller.

Hunter features 752 MI300A APUs spread across 188 liquid-cooled nodes. Each APU integrates 24 Zen 4 CPU cores, six CDNA 3 GPU tiles, and 128GB of unified HBM3 memory. These are complemented by 512 additional Epyc processors, each with 32 cores.

This configuration delivers an estimated peak theoretical performance of 48.1 petaFLOPS for double-precision scientific workloads—nearly double the performance of its predecessor, Hawk. However, Hunter is also optimized for AI tasks, with BF16 and FP8 data types enabling AI performance ranging from 736 petaFLOPS to 1.47 exaFLOPS depending on the workload’s precision requirements.

Under a public-private partnership, German AI startup Seedbox.ai is already using Hunter to train large language models (LLMs) in 24 European languages.

“The rapid development of AI and an increased focus on sustainability in supercomputing mean that high-performance computing is currently going through an exciting transformative period,” said Michael Resch, director of HLRS, in a statement.

Efficiency Meets Performance

Hunter introduces a “dynamic power capping” feature developed with HPE to improve energy efficiency. This system continuously monitors workloads and adjusts power distribution accordingly, reducing energy use by up to 20 percent without compromising performance. Early testing on the Hawk supercomputer demonstrated the potential of this feature.

A Stepping Stone to Herder

While Hunter’s 48 petaFLOPS won’t secure a spot among the Top500’s top 10 supercomputers, it serves a different purpose. As a transitional system, Hunter is designed to prepare researchers for larger GPU- and APU-based architectures in anticipation of the Herder supercomputer.

Herder, also to be built by HPE’s Cray division, is expected to deliver several hundred petaFLOPS of double-precision performance. Together, the systems represent a €115 million investment in advancing Germany’s HPC capabilities.

Construction of Herder’s infrastructure is expected to begin soon, with installation scheduled for completion in 2027.

Previous articleBest GTA 5 Cheats for PS4 & PS5: Your Ultimate Guide
Next articleSupreme Court Upholds Law to Ban TikTok in US

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.