DKRZ awards Bull EUR 40m climate supercomputer contract

DKRZ has awarded Bull a €40 million contract to build Germany's next climate research supercomputer, including five years of support for the new HLRE-5 system.

The machine is due to enter service in early 2028 at the German Climate Computing Centre in Hamburg. It will replace the current Levante system for a new phase of climate modelling that combines traditional simulations with artificial intelligence.

HLRE-5 will be built on Bull's BullSequana XH3500 platform. It is designed to run two global climate simulations at 5 km resolution at the same time while cutting the energy required per simulated year by about 75% compared with Levante.

The project extends a long-running partnership between the computing centre and Bull on energy use in large-scale research computing. The GPU extension of Levante ranked third in the Green500 lists published in late 2025 and mid-2026, while other Bull systems took the top two spots.

That focus on energy use is central to the investment, as climate centres face pressure to expand computing output without increasing electricity demand at the same pace. Kilometre-scale modelling requires far more computing resources than lower-resolution systems, particularly when researchers run multiple scenarios and combine them with growing volumes of observational data.

"The new system will allow us to drastically reduce the energy needed for our users' climate simulations. For a centre that exists to study the climate, that is not a side benefit - it is part of the job," said Dr. Hendryk Bockelmann, Head of DKRZ's Applications Department.

System design

The new machine will have two main partitions: a GPU partition using NVIDIA GB200 Grace Blackwell Superchips for the main simulation and AI work, and a CPU partition based on NVIDIA Vera CPUs.

The CPU side also introduces Arm architecture into DKRZ's core system for the first time. According to the centre, that should improve energy efficiency for workloads that do not run on GPUs.

All nodes in the system will be linked through NVIDIA Quantum-2 InfiniBand networking, providing inter-node communication at 400 Gbit/s. That is a standard requirement for tightly coupled climate models operating at kilometre-scale resolution, where very large volumes of data must move quickly between processors.

A member of the team behind the new system also highlighted the role of previous GPU optimisation work. "We are grateful for NVIDIA's longstanding support in optimizing ICON for GPUs, which also contributed to us winning the 2025 ACM Gordon Bell Prize for Climate Modelling. We look forward to continuing this collaboration on HLRE-5," said Engels.

Storage shift

The storage design reflects a wider shift in research computing from simply adding more capacity to managing data more selectively. The system will use IBM Storage Scale System 6000, with 60 petabytes of disk storage, throughput of up to 340 gigabytes a second, and a dedicated 500-terabyte flash tier.

The storage arrangement is intended to use tiering and data placement to match different research demands more closely. Flash storage is expected to support AI workloads, while the larger disk layer will hold broad volumes of climate data and simulation output.

This matters because climate science produces large, persistent datasets that are often reused across many research projects. AI methods can add further pressure by requiring repeated high-speed access to selected training and inference data.

Heat reuse

Cooling is another part of the project. As with earlier systems, HLRE-5 will use Bull's direct liquid cooling technology, which circulates warm water in a closed loop through computing nodes rather than relying on conventional air cooling.

DKRZ already sends part of the waste heat from its systems to the neighbouring chemistry department at the University of Hamburg. With HLRE-5, the cooling infrastructure will be upgraded so that a larger share of waste heat can be reused, and the university plans to extend that supply to more buildings on and around the Climate Campus.

Funding for the system is mainly being provided through a long-term agreement involving the Helmholtz Association, the Max Planck Society, and the City of Hamburg. Additional funding from the Max Planck Society and the Federal Ministry of Research, Technology and Space increased the planned size of the system.

"We are proud to continue our long-standing collaboration with DKRZ, one of Europe's leading climate computing centres. HLRE-5 builds on years of joint innovation in energy-efficient supercomputing and demonstrates how advanced digital infrastructure can support excellent climate research," said Dr. Martin Matzke, Head of Northern and Central Europe, Bull.