Cold Plate Liquid Cooling:
Principle:
A metal (usually copper or aluminium) cold plate is installed on key components (CPU, GPU, memory, etc.) that generate a lot of heat. The coolant flows in micro-channels inside the cold plate and directly absorbs the heat from the chip. Other components that generate relatively low heat (e.g., hard drives, power supplies, some motherboard components) still rely on air to dissipate heat.
Cooling Circuit: Coolant circulates between the CDU or cabinet-level heat exchanger inside the cabinet and the server cold plate. the CDU/heat exchanger then transfers the heat to a secondary cooling water circuit (e.g., chilled water) at the server room level.
Advantages:
Relatively minor modifications to the server (mainly in the radiator and pipework connections), good compatibility.
Can be deployed gradually, mixed with air-cooled presence.
Relatively mature technology and better industry chain.
Applicable scenarios: Moderately high power density (e.g., 20-50kW/cabinet), high compatibility requirements for retrofitting existing facilities, or scenarios as a transitional solution.
Immersion Liquid Cooling:
Principle:
The entire server or server node is completely submerged in a non-conductive coolant. The coolant is in direct contact with all electronic components and absorbs heat.
Advantages:
Ultimate heat dissipation: Direct contact with minimal thermal resistance for maximum cooling efficiency, easily handling very high power densities (>100kW/cabinet).
Complete elimination of fans: No fans inside the servers, very low noise, further reducing energy consumption.
High density deployment: Servers can be tightly packed for high space utilisation.
Simplified infrastructure: Dramatically reduced or even eliminated server room air conditioning, air ducts, raised floors, etc.
High reliability: Insulation from oxygen and humidity reduces corrosion and dust contamination and improves equipment reliability.
Efficient use of natural cooling sources: Especially suitable for combining with natural cooling methods such as dry coolers to achieve extremely low PUE throughout the year.
Applicable scenarios: Newly constructed large-scale computing centres, scenarios with strong demand for extreme PUE and ultra-high density (e.g. large-scale AI training clusters, supercomputing centres).