As the emerging business models and advanced customer experience driving an explosion of data at the era of 5G, data center has been evolving with increasingly powerful AI (Artificial Intelligence) computing capability for enhancing CSP (cloud service provider) cloud service by data analytics in data centric world. High-performance high-power computing, storage and networking are backend behind this. Meanwhile, having data center greener for energy efficiency is mandatory design target for modern data center, measured with PUE. Advanced thermal solution is one of crucial technology to continually address the demand of higher performance computing at lower PUE.
D2C (Direct to Chip) liquid cooling, as a typical advanced thermal solution, has been accepted by the industry as an effective cooling solution for high density datacenters. JD.com has years’of liquid cooling engineering design experience from rack level to data center level. This presentation introduces best known method (BKM) from JD.com engineering practice and deployment data. It also focuses on system level liquid cooling design related ingredients such as CPU cold plates, memory cold plates, leakage detection, quick disconnects, etc. Moreover, this presentation details a JD.com own advanced design methodology by engineering practice data analytics, optimal for CPU cold plate design. Also, whole-loop leakage detection has been considered as one of the pain points in the industry, this presentation will also discuss the implementation of whole-loop leakage detection in a JD datacenter.
