If you are an operator dealing with massive heat loads from AI chips—this project developed a two-phase cooling system that handles heat fluxes like the Nvidia H100's 70 W/cm2. It recovers 50% of waste heat to create more cooling power, slashing electricity bills.
Self-Cooling Energy Efficient Systems for High Performance Computing Data Centers
Imagine a computer that uses its own waste heat to power its own air conditioner. Instead of using massive, power-hungry fans and chillers, this system uses special 3D-printed materials and liquids to soak up heat and turn it back into cooling. It's like a smart thermos that keeps the processor cool by recycling the heat it generates.
What needed solving
High-performance computing chips generate extreme heat that traditional air and water cooling cannot handle efficiently. This leads to massive energy waste and expensive cooling infrastructure costs.
What was built
A system of adsorption chillers using 3D-printed heat exchangers and composite adsorbents. It includes a two-phase cooling loop and an AI-assisted management tool for processor optimization.
Who needs this
Who can put this to work
If you are a manufacturer dealing with space constraints in small data centers—this project developed rack-integrated adsorption chillers for rear doors. This allows high-power processors to be cooled without needing separate, bulky external cooling infrastructure.
If you are a provider dealing with overheating in remote HPC installations—this project developed an active management system that optimizes processor tasks based on cooling capacity. This prevents thermal throttling and extends hardware lifespan.
Quick answers
What is the estimated cost or price of the system?
Based on available project data, specific pricing or cost-per-unit is not provided; the project focuses on technical development and efficiency gains.
Is this technology ready for industrial scale?
The project is currently developing new designs for adsorption chillers and 3D-printed heat exchangers. It targets two scales: large HPC supercomputers and small individual racks.
How is the IP and licensing handled?
Based on available project data, specific licensing terms are not mentioned, though the project involves 8 industrial partners who likely contribute to and benefit from the IP.
How does it integrate with existing hardware?
It integrates via two-phase cooling for processors and an interoperable control system based on the EAR platform, including a rear-door solution for racks.
What is the project timeline for deployment?
The project period runs from 2024-07-01 to 2027-06-30, suggesting the technology will be refined through 2027.
Who built it
The consortium is heavily industry-driven with a 67% industry ratio, comprising 8 SMEs and 4 other partners across 5 countries. This strong industrial presence, combined with only one university and two research centers, suggests a high priority on commercial viability and practical application over pure academic research.
Contact IDP INGENIERIA Y ARQUITECTURA IBERIA SL in Spain
Talk to the team behind this work.
Contact us to find partners for the 2027 commercial rollout of these cooling technologies.