The German Climate Computing Center (DKRZ) has selected Bull, the high-performance computing arm of Atos, to supply its next-generation supercomputer for earth system research, the HLRE-5, a system designed to run two simultaneous global climate simulations at 5-kilometer resolution while slashing energy consumption per simulated year by roughly 75 percent compared with the center's current Levante system.
The contract, announced on October 26, 2026, builds on a long-standing partnership between DKRZ and Bull in energy-efficient supercomputing. The current GPU extension of the Levante system ranked third among the world's most energy-efficient supercomputers in the November 2025 and June 2026 editions of the Green500 list, with the top two positions also held by Bull-built systems. The new system, expected to be available to users in early 2028, represents a total investment of 40 million euros, including a five-year support agreement, funded primarily by the Helmholtz Association, the Max Planck Society and the City of Hamburg under the long-term financing arrangement for HPC systems at DKRZ. Additional funds from the Max Planck Society and the Federal Ministry of Research, Technology and Space under Germany's Hightech Agenda enabled a significant expansion of the system's scale.
"With the new system, we can drastically reduce the energy required for our users' climate simulations. For a data center whose mission is to support climate research, that is not an added benefit — it is part of our mandate," said Dr. Hendryk Bockelmann, head of applications at DKRZ. Dr. Martin Matzke, head of Northern and Central Europe at Bull, said the company was proud to continue its long collaboration with "one of Europe's leading climate computing centers," adding that HLRE-5 demonstrates how modern digital infrastructure can support excellent climate science.
HLRE-5 comprises two partitions: a GPU partition based on NVIDIA GB200 Grace Blackwell superchips that provides the bulk of the system's simulation and AI performance, and a CPU partition built on NVIDIA Vera CPUs that brings the ARM architecture into DKRZ's core system for the first time and substantially improves energy efficiency for workloads not suited to GPUs. All compute nodes are interconnected via NVIDIA Quantum-2 InfiniBand, enabling low-latency communication at data rates of 400 Gbit/s — a requirement for global climate models at kilometer-scale resolution. DKRZ noted that NVIDIA's long-standing support in optimizing the ICON climate model for GPUs contributed to its winning the 2025 ACM Gordon Bell Prize for Climate Modelling, and that the collaboration will continue on HLRE-5.
For data storage, DKRZ is shifting from capacity as the primary sizing metric toward efficient resource use through intelligent data placement across storage tiers. At the center is IBM's latest Storage Scale System 6000, which combines 60 petabytes of disk storage with throughput of up to 340 gigabytes per second and a dedicated 500-terabyte flash tier. Working closely with IBM, DKRZ will analyze and further optimize advanced functions for on-demand data distribution across storage hierarchies and for allocating storage resources to different research tasks. The flash tier is designed to meet the rapidly growing demands of AI workloads, giving the research community a foundation for both HPC and next-generation AI-driven science.
As with previous generations, DKRZ will rely on Bull's patented direct liquid cooling technology, in which warm water circulates through compute nodes in a closed loop and removes heat far more efficiently than air cooling. For several years, DKRZ has supplied part of this waste heat to the neighboring chemistry department at the University of Hamburg; with the installation of HLRE-5, the cooling infrastructure will be modernized so that a larger share of waste heat is available for reuse, and the university plans to extend that use to additional buildings on and around the Climate Campus.
