Jan. 16, 2025 — A celebration held today at the High-Performance Computing Center of the University of Stuttgart (HLRS) greeted the start of operation of HLRS’s newest supercomputer, called Hunter. Designed and manufactured by Hewlett Packard Enterprise (HPE), Hunter will offer a world-class infrastructure for large-scale simulation, artificial intelligence, and data analytics applications in science, industry, and the public sector.
Hunter is also designed with sustainability in mind, making use of 100% fanless direct liquid cooling system architecture and a unique dynamic power capping approach to maximize its energy performance. Based on energy efficient AMD Instinct Accelerated Processing Units (APUs), which combine CPU cores and GPUs in an integrated APU-accelerated architecture, Hunter also marks the beginning of a new chapter in the history of supercomputing in Stuttgart.
Users of Hunter will address complex problems in engineering, weather and climate modeling, biomedical research, and materials science, among other fields. Hunter will also provide secure access to powerful, secure high-performance computing (HPC) and AI resources for industry and for public sector agencies.
“Hunter offers scientists at the University of Stuttgart and across Germany a future-proof infrastructure for AI-based simulations and high-performance computing of a new quality,” said Prof. Peter Middendorf, Rector of the University of Stuttgart. “Hunter also benefits the entire ecosystem of our university with its global players, its strong medium-sized companies, and its growing start-up scene.”
“The rapid development of AI and an increasing focus on sustainability in supercomputing mean that high-performance computing is currently going through an exciting, transformative period,” said Prof. Michael Resch, Director of HLRS. “With Hunter, our user community gains a state-of-the-art infrastructure that will support them in navigating this changing HPC landscape and enable them to remain competitive at the frontiers of scientific discovery and industrial innovation.”
Funding of 15 million Euros
The total cost of Hunter was 15 million Euros. The Baden-Württemberg Ministry for Science, Research, and Art provided half of the funding in conjunction with its high-performance computing/data intensive computing (HPC/DIC) strategy. Germany’s Federal Ministry for Education and Research provided the second half in the context of the SiVeGCS project. Financing was facilitated by the Gauss Centre for Supercomputing (GCS), the alliance of Germany’s three national supercomputing centers.
“A high-performance computer like the new Hunter is not simply a technical innovation — more importantly, it is necessary for the future capabilities of our country and a key to progress in science and industry. This in turn is a foundation for our prosperity,” said Petra Olschowski, Minister of Science, Research and Art of the State of Baden-Württemberg. “Baden-Württemberg leads Europe in the fields of supercomputing and artificial intelligence and is internationally competitive. HLRS at the University of Stuttgart plays a key role in this. Hunter will make it possible to model the next generation of safer, quieter, and more environmentally friendly aircraft like helicopters, flying taxies, and drone-supported delivery services. The additional computing performance is also relevant for progress in quantum mechanics, climatology, astrophysics, and energy research.”
Increased Productivity, Better Energy Efficiency
Hunter is based on the same HPE Cray Supercomputing EX4000 architecture used in the world’s three verified exascale systems. Each of its 136 nodes is equipped with four HPE Slingshot high-performance interconnects. The machine leverages HPE Cray Supercomputing Storage Systems E2000, which is purpose-engineered to meet the demanding input/output (I/O) requirements of large-scale supercomputers, and HPE Cray Supercomputing Programming Environment Software, which offers a comprehensive set of tools for developing, porting, debugging, and tuning applications at scale. HPE Performance Cluster Manager is also used for system health monitoring and management as well as power management. In addition, the entire system architecture of Hunter makes use of 100% fanless direct liquid cooling, a first-of-its-kind technology pioneered by HPE that is designed to enhance the energy and cost efficiency of large-scale AI deployments.
“From the very beginning, HLRS has uniquely supported both cutting-edge science and industrial applications – as a result, HLRS has become an important competitive factor for Baden-Württemberg and beyond,” said Heiko Meyer, Executive Vice President and Chief Sales Officer, Hewlett Packard Enterprise. “We are proud to help advance HLRS’s mission by taking a big step into the future of supercomputing and AI with Hunter, and subsequently with Herder. This will open up completely new opportunities for foundational research and industrial product innovation – while at the same time achieving breakthroughs in energy-efficient supercomputing.”
With a theoretical peak performance of 48.1 Petaflops (48.1 quadrillion floating point operations per second), Hunter’s speed is nearly double that of HLRS’s previous flagship supercomputer, called Hawk. At the same time, Hunter represents an important shift in computing architecture for the center. Whereas past systems have relied primarily on large numbers of CPUs to achieve high performance, Hunter is based on the AMD Instinct MI300A accelerated processing unit (APU), which combines CPUs, GPU accelerators, and high bandwidth memory in a single package. At the same time that it enables higher performance, this system architecture means that Hunter consumes 80% less energy at peak performance than Hawk.
“The AMD Instinct MI300A APUs are driving innovation by delivering leadership performance and efficiency for critical workloads at the convergence of HPC and AI,” said Brad McCredie, Senior Vice President, Data Center Engineering, AMD. “We are pleased with the longstanding successful collaboration with HLRS and excited to power Hunter as one of the most advanced supercomputers in Germany using AMD Instinct MI300A APUs. Together with the teams at HPE and HLRS, we are proud to be paving the way toward the exascale era.”
Hunter will also implement a dynamic power capping software functionality developed by HPE in collaboration with HLRS to maximize computational productivity and energy efficiency. This functionality will continuously monitor the applications running on Hunter, optimize power distribution to each application based on its individual power requirements, and ensure that overall energy consumption does not exceed a predetermined power limit. After dynamic power capping was implemented on HLRS’s Hawk system, a study showed that the system software reduced overall power consumption of applications by about 20% without significant losses in performance.1 The study demonstrated that the overall savings on Hawk were similar to the annual power consumption of 1,500 single family homes.2 HPE and HLRS will develop the method further on Hunter to improve its effectiveness on GPUs.
More Power for Artificial Intelligence
The adoption of an architecture based on CPUs and APU accelerators will also make Hunter a powerful system for artificial intelligence. Although Hunter will continue to support traditional applications of high-performance computing, it also offers the potential to expand HLRS’s user community to include data scientists and artificial intelligence specialists interested in developing customized large language models, deep learning projects, and complex data analytics.
Demonstrating the potential of Hunter for supporting new kinds of artificial intelligence applications, Stuttgart-based AI startup Seedbox.ai has already begun using the new system during its test phase to train new large language models in 24 European languages. The company will make these models available on an open-source basis to enable AI-based innovation — for example, to accelerate the implementation of multilingual autonomous agents in everyday working life. European users will benefit not only from the wide range of languages for which the models have been optimized, but also from significantly improved energy efficiency. As part of a partnership with HLRS signed in 2024, Seedbox will continue using Hunter to develop scalable, sustainable, and secure “Made in Germany” AI technologies.
In addition, the combination of CPUs and GPUs in a single package will make it easier to develop and run new kinds of converged computing workflows that combine simulation, data analysis, and artificial intelligence in innovative ways. Simulation could be used on Hunter, for example, to produce synthetic data sets for the training of AI algorithms. Conversely, artificial intelligence could be integrated into simulation workflows to accelerate computationally expensive codes.
Hunter will be a backbone for artificial intelligence in the initial phase of HammerHAI, a German AI Factory announced by the EuroHPC Joint Undertaking in late 2024. Once the project launches later this year HammerHAI will develop a portfolio of solutions and services that will lower the barriers that currently limit European use of AI. In 2026 a new EuroHPC Joint Undertaking AI-optimized supercomputer will arrive at HLRS, providing large-scale AI computing performance for German and European science and industry.
A Stepping Stone to the Next Level of Performance
Hunter is conceived as a transitional system that will prepare the way for HLRS’s next supercomputer, called Herder, which is being planned for installation by HPE in 2027. With a top speed of several hundred petaflops, Herder will constitute a major jump in peak performance over Hunter. Because Hunter will leverage a similar GPU-accelerated approach to Herder, it will offer HLRS’s user community the opportunity to prepare their application codes to take full advantage of the massive increase in computational power that will soon be available.
As was also discussed at Hunter’s inauguration, Herder will be housed in a new building and power facility whose construction will soon begin at HLRS. The facility, called HLRS III, will use sustainable materials, be outfitted with photovoltaic panels, and distribute heat generated by Herder to warm other buildings on the University of Stuttgart’s Vaihingen campus. In recognition of its sustainable infrastructure planning, HLRS was named the winner of the Data Center Strategy Award for “Transformation” in October 2024.
Source: Christopher Williams, HLRS
This post was originally published on here
