Stuttgart, December 7, 2023 – Today, Lenovo announced the expansion of the High-Performance Computing (HPC) system “LISE” at the Zuse Institute Berlin (ZIB) – an interdisciplinary research institution for applied mathematics and data-intensive HPC. This expansion provides researchers with the necessary power to execute compute-intensive applications such as artificial intelligence and machine learning. The focus of the expansion has been on improving energy efficiency, utilizing innovative technologies, including Lenovo Neptune liquid cooling, and ensuring seamless integration into the existing system by HPC partner pro-com.
Technology and Sustainability
Depending on workload, inlet temperature, and ambient temperature, the system’s efficiency is in the range of 85-90%. By employing the Lenovo Extreme Heat Recovery solution with water-cooled power supplies and insulated, water-cooled rear doors of the HPC racks, efficiency is further enhanced, with over 97% of the heat directed into the hot water circuit instead of the data center. This enhances the data center’s energy balance, eliminating the need for additional room cooling. A feasibility study is currently underway to explore the reuse of waste heat for the district heating network or local use at ZIB.
“For the Zuse Institute Berlin, the energy-efficient operation of its powerful HPC resources has been of particular importance for years, standing on equal footing with high computing power and efficient operation,” added Thomas Steinke, Head of the Supercomputing Department at ZIB. “In terms of sustainability and energy cost development, we have chosen technological solutions from Lenovo that allow us to meet our requirements for both strong computing power through the latest processor generations and high energy efficiency of CPU and GPU platforms due to direct liquid cooling. We are pleased to have found strong partners in Lenovo and pro-com DATENSYSTEME GmbH for the implementation of our plans.”
The LISE system was commissioned at the end of 2019 and expanded this year with GPU partitions from NVIDIA and Intel due to increased demand for GPU resources. The NVIDIA GPU partition consists of 42 computing nodes, each containing 4 NVIDIA A100/80 GPUs. The Intel GPU partition, on the other hand, comprises 8 computing nodes, each equipped with 4 Intel MAX GPUs (formerly known as “Ponte Vecchio”). Both partitions utilize Lenovo hardware. The NVIDIA GPU partition has been accessible to users since the end of March, while the Intel partition is currently being prepared for operation. Both GPU partitions are planned to be utilized for four to five years and will be expanded with new heterogeneous processor architectures from 2025. Furthermore, the installation of an energy-efficient CPU partition with 168 nodes is scheduled for the end of 2023/early 2024, utilizing the Lenovo ThinkSystem SD665 V3 with AMD EPYC processors of the 4th generation and integrated Neptune liquid cooling.
“We are pleased that the Zuse Institute Berlin, together with pro-com DATENSYSTEME GmbH, has chosen Lenovo for the expansion of their HPC cluster ‘LISE’ to further advance research in Germany,” commented Andreas Thomasch, Director HPC & AI DACH, France, UKI at Lenovo. “This choice underscores our commitment to technological progress through the warm-water-cooled deployment of the latest Intel GPUs and NVIDIA GPUs, as well as AMD and Intel CPUs, combined with our commitment to sustainable and efficient solutions in the field of High-Performance Computing. Through our innovative Neptune liquid cooling, we enhance the computing power of the cluster for even faster insights for scientists while simultaneously reducing environmental impact.”
Applications
The LISE system at ZIB supports diverse applications from various scientific disciplines, particularly in life sciences, chemistry and materials science, earth system science, and engineering (CFD). GPU resources play a central role in projects employing Artificial Intelligence (AI) and Machine Learning (ML) methods in simulations and method development. Two examples that combine traditional simulation methods with AI approaches are drug development and the study of the dynamics of protein structures. The first project involves an in-silico search for new opioids without harmful side effects, coupling complex atomistic simulations with AI methods to effectively identify candidates and their mechanisms of action. In another project in collaboration with the Free University of Berlin, new atomistic-based protein models are refined using AI approaches to simulate important structure-dependent protein functions that develop over an extended timescale.
Funding and Network
ZIB is one of nine NHR (National High-Performance Computing) centers in the National High-Performance Computing Consortium, aiming to further expand high-performance computing capacities for research and development in Germany. The collaboration and networking of HPC centers in the NHR consortium increase efficiency and access to powerful computing infrastructure for scientists and researchers throughout Germany. The HPC Cluster LISE plays a significant role in this network, helping establish Germany as an innovation force in the field of high-performance computing. LISE is now being expanded as part of this program with financial support from federal and state governments.
