FUTURE DATA CENTERS: LIQUID IMMERSION COOLING INNOVATION TO WITHSTAND AI HEAT
Journal of Computer Science Advancements · 17.08.2025 · 3 (4) · 220-234
DOI: 10.70177/jsca.v3i4.3333 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Computer Science Advancements, online erschienen am 17.08.2025. Verfasst von Aom Thai und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).
Abstract (Originalsprache)
The exponential escalation of computational density required by modern Artificial Intelligence (AI) and Large Language Models has pushed traditional air-cooled data center infrastructures to their thermodynamic limits. This study investigates the efficacy of single-phase liquid immersion cooling as a transformative solution to manage the extreme thermal flux of next-generation AI accelerators. Adopting a quantitative experimental design, we benchmarked a high-density GPU cluster submerged in a proprietary dielectric fluid against a standard forced-air baseline under intensive MLPerf training workloads. The research focused on evaluating key performance indicators, including Power Usage Effectiveness (PUE), processor junction temperatures, and total energy consumption over a 168-hour stress test. Results demonstrate that the immersion architecture achieved a near-ideal PUE of 1.04, representing a 34% efficiency improvement over the air-cooled control group. Furthermore, the liquid medium maintained GPU core temperatures 20°C lower than the baseline, effectively eliminating thermal throttling events and enhancing computational stability. The study concludes that shifting from aerodynamic to hydrodynamic cooling is not merely an efficiency upgrade but a physical prerequisite for the sustainable scaling of exascale AI infrastructure, offering a viable pathway to decarbonize the expanding digital economy.
Zitieren
APA
Thai, A., Krit, P., & Lek, S. (2025). FUTURE DATA CENTERS: LIQUID IMMERSION COOLING INNOVATION TO WITHSTAND AI HEAT. Journal of Computer Science Advancements, 3(4), 220-234. https://doi.org/10.70177/jsca.v3i4.3333
Vancouver
Thai A, Krit P, Lek S. FUTURE DATA CENTERS: LIQUID IMMERSION COOLING INNOVATION TO WITHSTAND AI HEAT. Journal of Computer Science Advancements. 2025;3(4):220-234. doi:10.70177/jsca.v3i4.3333
BibTeX
@article{Thai2025,
author = {Thai, Aom and Krit, Pong and Lek, Siri},
title = {FUTURE DATA CENTERS: LIQUID IMMERSION COOLING INNOVATION TO WITHSTAND AI HEAT},
journal = {Journal of Computer Science Advancements},
year = {2025},
volume = {3},
number = {4},
pages = {220--234},
doi = {10.70177/jsca.v3i4.3333}
}
EUVIMED