Rising heat density in GPU-accelerated servers has amplified the risk of incipient fires and thermally induced hardware degradation in modern data centers. This study proposes a dual-safety hybrid that integrates (i)thermo-responsive, power-independen...
Rising heat density in GPU-accelerated servers has amplified the risk of incipient fires and thermally induced hardware degradation in modern data centers. This study proposes a dual-safety hybrid that integrates (i)thermo-responsive, power-independent suppression capsules with (ii) a direction-steerable rotary cooling unit. Unlike conventional solutions that depend on continuous power, sensors, and controllers, the proposed architecture preserves functionality under blackout conditions, removing single points of failure. Experiments on a 2U server equipped with an RTX A6000 GPU (300 W TDP) reproduced high-load and thermal-surrogate conditions and employed thermocouples, high-speed videography, and infrared thermography for quantitative assessment. The system reduced average GPU temperature by ΔT ≈ 9.2 °C (≈11.2%), lowered fan power from 13 W to ~9 W (−30.7%), and—under fire-surrogate heating—triggered capsule discharge within ≤0.45 s, dispersing a non-conductive, non-corrosive fluorinated-ketone agent over a radius ≥ 25 cm. These results demonstrate that a power-agnostic, self-actuating cooling–suppression mechanism can mitigate fire risk and improve thermal/energy performance in data-center environments, supporting near-term applicability to unattended or power-unstable scenarios.
Keywords: data-center safety, GPU server, thermo-responsive suppression, passive actuation, rotary cooling, energy efficiency