IT-power feedforward halves pump energy in liquid-cooled data centres

Abstract Pump work in liquid-cooled data centres scales with the cube of flow rate, so the choice of controlled variable sets the achievable energy floor. Existing pump controllers regulate temperature — the result of the thermal balance — although the heat load (IT power), its cause, is directly readable from power distribution units. Here we show that replacing temperature feedback with a closed-form feedforward from IT power reduces pump work by 56.2% and peak junction temperature by 5.14 °C, against a temperature controller tuned over 90 parameter sets, in a 90 kW rack under a 30%→100% step. Conversely, a 300 s look-ahead increases pump work by 45.5% without changing peak temperature, because required flow is an instantaneous function of power. Inter-rack reallocation is likewise infeasible: each lower bound is set by its own load and cannot be transferred. The gain lies in the observed variable, not in algorithmic sophistication.

Authors

Publication Details

Journal
Research Square
Published
2026-10-29
DOI
https://doi.org/10.21203/rs.3.rs-11480780/v1
Primary Topic
Energy Efficiency in Computing
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
preprint

IT-power feedforward halves pump energy in liquid-cooled data centres

Yang Liu
3 links
Research Square
Energy Efficiency in Computing
preprint

IT-power feedforward halves pump energy in liquid-cooled data centres

Yang Liu
preprint en

Abstract

Abstract Pump work in liquid-cooled data centres scales with the cube of flow rate, so the choice of controlled variable sets the achievable energy floor. Existing pump controllers regulate temperature — the result of the thermal balance — although the heat load (IT power), its cause, is directly readable from power distribution units. Here we show that replacing temperature feedback with a closed-form feedforward from IT power reduces pump work by 56.2% and peak junction temperature by 5.14 °C, against a temperature controller tuned over 90 parameter sets, in a 90 kW rack under a 30%→100% step. Conversely, a 300 s look-ahead increases pump work by 45.5% without changing peak temperature, because required flow is an instantaneous function of power. Inter-rack reallocation is likewise infeasible: each lower bound is set by its own load and cannot be transferred. The gain lies in the observed variable, not in algorithmic sophistication.

Research Square
Affordable and clean energy, Industry, innovation and infrastructure
Energy Efficiency in Computing
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon