LRZ Decommissions CoolMUC-3 Supercomputer After Almost Decade of Use

Why It Matters

The project validates direct hot‑water cooling as a scalable, energy‑efficient solution for future high‑performance computing installations, influencing sustainability standards across the industry.

Key Takeaways

• CoolMUC‑3 operated 2017‑2026, delivering 400 TFLOPS.
• First HPC with near‑100% hot‑water cooling.
• Fanless design transferred 97% heat to water.
• Served major Bavarian universities for fluid dynamics.
• Decommissioning emphasized reuse and environmental responsibility.

Pulse Analysis

The retirement of CoolMUC‑3 marks the end of a pioneering era in German high‑performance computing. Launched in 2017, the system’s 400 teraflops capacity supported advanced fluid‑dynamic simulations for institutions such as the Technical University of Munich and LMU. Its decade‑long operation demonstrates the durability of Intel Xeon Phi architectures in research environments, while the structured decommissioning process underscores LRZ’s commitment to responsible hardware lifecycle management.

At the heart of CoolMUC‑3 was Megaware’s ColdCon liquid‑cooling technology, the first HPC system to employ near‑100 percent direct hot‑water cooling with inlet temperatures of 40 °C. By eliminating fans and channeling 97 percent of waste heat into a water circuit, the cluster achieved silent operation and dramatically reduced power consumption for cooling. This approach aligns with broader data‑center trends toward higher inlet temperatures, lower PUE (Power Usage Effectiveness) ratios, and the reuse of waste heat for district heating or other industrial processes.

The successful decommissioning—recycling viable components and disposing of the remainder responsibly—provides a template for sustainable HPC turnover. As the industry grapples with the environmental impact of ever‑larger supercomputers, CoolMUC‑3’s legacy offers concrete evidence that direct liquid cooling can be both technically viable and ecologically sound. European research centers are likely to adopt similar designs, accelerating the shift toward greener, more cost‑effective supercomputing infrastructures.