This activity aims at pioneering, groundbreaking techniques to cool down Data Center (DC) servers before converting the recuperated heat into reusable energy. This will be achieved by optimally cooling the DC CPUs very close to their hot spots. This allows:

a) effective heat extraction from CPUs, at up to 3x-higher temperature of coolant (75oC) than conventional solutions for subsequent energy recuperation

b) improvement of CPUs performance

c) extentions of servers lifetime

The solutions will be eventually scaled up to 200 decommissioned servers in the EPFL datacenter, plus a liquid distribution system, to extract 50 kW of heat power.

Technological roadmap

For the implementation of this activity, the following steps will be undertaken:

Phase 1. Optimization and validation for a full server rack

During Phase 1 (2023-2026), the focus will be on the validation of the proposed solution at the lab scale.

1. Topology-based solver to design the optimal microfluidic network
2. Design of packaging, server, system integration, and test facility
3. Manufacturing cold plates and integration on server rack demonstrator

Phase 2. Scale up, design, development, and validation of a multi-server rack

During Phase 2 (2026-2028), the focus will be on the scaling up of this solution in collaboration with the EcoCloud industrial partners, such as Microsoft, Intel, HPE, and Meta.

1. Scale up the technology to 400 CPUs (retrofitted from SCITAS/EPFL decommissioned servers)
2. Construct a liquid-distribution system connecting all servers to extract 50kW of heat power.
3. Provide high-temperature coolant (75oC) to the following activities of the project (ORC and heat sharing) to reuse the extracted heat.

Expected results

  • Significant energy savings on the EPFL data center
  • Scaled-up system efficiently extracting 50 kW of heat power and high-temperature coolant (75oC).