Key Secondary-side Components Build Complete Cold Plate Liquid Cooling Architecture
The secondary cooling loop takes surplus cooling capacity from the primary circuit to dissipate heat for core heat-generating components inside racks, including CPUs, GPUs, memory modules, hard drives and power supplies. Its main components cover CDU, site distribution pipelines and liquid cooling cabinets equipped with cold plates, quick disconnect fittings and manifolds, forming the core operating framework of a standard liquid cooling data center. This article systematically introduces all core liquid cooling parts of the secondary loop.
Cold plates serve as direct heat-capturing components, manufactured from high thermal-conductivity copper or aluminum with elaborate microchannel structures. Water-based coolant flows inside to carry away heat from heated chips. Two critical indicators evaluate cold plate performance: pressure drop and thermal resistance. Excessive internal flow resistance slows down coolant speed and reduces heat exchange efficiency, while lower thermal resistance delivers superior cooling capacity, which is essential for stable operation of water cooled data center.
Apart from structural optimization of microchannels, phase-change technology greatly improves cold plate performance. Coolant with low boiling point absorbs abundant latent heat via vaporization and keeps consistent cooling for multiple processors. Two-phase cold plates are the future upgrade direction of single-phase products with better cost performance than immersion cooling, gradually becoming an ideal matching component for mainstream water cooled server. Restricted by unstable gas-liquid two-phase flow control and insufficient unified industrial standards, domestic large-scale mass production of phase-change cold plates remains limited currently.
Manifold works as the fluid distribution hub connecting CDU and cold plates of liquid cooled server. Consisting of paired supply and return pipelines plus multiple branch ports, it enables closed coolant circulation: fluid pumped out from CDU is split to each cold plate, collects waste heat and flows back to CDU. Made of seamless welded 316L stainless steel for high strength and corrosion resistance, manifolds support customized branch layout according to actual cabinet space and server quantities.
Quick disconnect fittings connect manifolds and servers with matched male and female terminals integrated with sealing rings, springs and locking structures, enabling drip-free plugging under pressurized wet condition. Divided into manual plug-in and blind-mate types, blind-mate connectors automatically lock during server rack-in without manual operation. First commercialized by Huawei in full-rack products in 2019 and adopted on NVIDIA GB200 NVL72, blind-mate fittings continuously expand market share alongside mature CDU liquid cooling system deployment.
Coolant pipelines act as fundamental delivery carrier for fluid and heat transfer, split into flexible hoses and rigid metal pipes. Hoses are commonly made of EPDM, FEP or PTFE, while stainless steel dominates metal pipeline materials. All selected materials feature outstanding corrosion resistance and leak tightness to satisfy long-running rigorous requirements of liquid cooling for AI GPU data center.
Leak detection ropes are essential safety accessories preventing costly hardware damage and data loss caused by conductive coolant leakage. Leak inspection falls into indirect sensor-based monitoring and direct physical detection via leak-sensing cables. Positioning and non-positioning leak ropes trigger alarming once two internal conductive cores are bridged by water, widely deployed to secure stable running of liquid cooling high density computing clusters against unexpected fluid leakage risks.
Common circulating coolants include ethylene glycol solution, propylene glycol solution and deionized water. Mainstream OEM brands adopt 20%~30% mixing proportion with 25% as the preferred standard; appropriate concentration balances antifreeze, microorganism inhibition and fluidity to guarantee stable heat transfer in standardized rack level liquid cooling projects.
Two mainstream industry delivery modes are available on the market. The integrated full-rack solution realizes unified design and assembly by a single supplier and has become mature commercial application. The decoupled delivery mode separates cabinets and liquid cooling servers under unified interface specifications from different manufacturers, which helps drive the formulation of industrial standards, yet inconsistent specifications and ambiguous after-sales responsibility still restrain its large-scale popularization.
An international liquid cooling authority, EXTRCOOL designs fully customized solutions for high-density AI and HPC data centers. Our premium thermal hardware includes dual-mode CDUs, manifolds, cold plates, rear-door heat exchangers, dry coolers, and containerized units. Field-proven for stable performance, wide interoperability, and easy scalability — enabling smooth one-stop full liquid cooling. Backed by deep expertise, we serve data center operators and enterprises across North America, Central Europe, the Middle East, East Africa, and Australia.
