Design and implementation of a fire-responsive cooling–suppression integrated system for mitigating fire risks in data-center GPU servers

Abstract

This study addresses rising fire risk and thermal stress in GPU-accelerated data-center servers by proposing a power-agnostic, self-actuating safety architecture. We integrate thermo-responsive, rupture-on-heat suppression capsules with a direction-steerable rotary cooling module and evaluate the hybrid on a 2U RTX A6000 server under realistic rack conditions. Instrumentation includes type-T thermocouples, high-speed videography, and infrared thermography. Blackout trials verify actuation without external power, sensors, or controllers. The steerable cooling reduced average GPU temperature by ΔT ≈ 9.2 °C (≈11.2%) and lowered fan power from 13 W to ~9 W (≈30.7%). Under a 200 ± 10 °C surrogate heat input, capsules discharged within ≤0.45 s, dispersing a non-conductive, non-corrosive clean agent over ≥25 cm radius; no electrical or corrosive damage was observed on proximal components. A heat-triggered, power-independent cooling–suppression hybrid can mitigate incipient fire risk while improving thermal and energy performance, removing the single-point-of-failure inherent to electrically actuated systems. The modular design and passive actuation suit unmanned or power-unstable environments (edge nodes, defense/finance servers), enable straightforward retrofit to existing chassis, and support compliance-oriented safety upgrades without dependence on facility power continuity.