What is Liquid Cooling?
Liquid Cooling for Today and Tomorrow
The data center industry is undergoing significant changes due to two major trends: the rising power needs of GPUs for services like large language models and generative AI, and growing concerns about the environmental impact of data centers. These developments are pushing data center owners and operators to move beyond traditional cooling solutions. Hybrid environments that combine air and liquid cooling are becoming more common, particularly in data centers focused on running generative AI and improving energy efficiency.​
Cooling the Next Generation of Computing
High performance computing (HPC) and data-intensive technology applications – like machine learning (ML), artificial intelligence (AI), generative AI, crypto mining, and high-definition video processing – can have server densities over 100kW, making the ROI for air cooling unattractive. Liquid cooling, with its higher thermal conductivity, provides significantly greater capacity to reject heat and cool the server densities that these technologies require.
​
Typical applications and use cases include artificial intelligence (AI), advanced cloud computing, video and graphics rendering, high performance computing (HPC), and any low latency, high processing applications.
Different types of Liquid Cooling
Direct-to-Chip
In a Direct-to-Chip solution liquid coolant is delivered directly to the hottest components, usually the GPU and/or CPU via a cold plate placed directly on the chip. The electric components are never in direct contact with the coolant and the liquid coolant is circulated through channels within the cold plates, absorbing heat directly from the components. The heated coolant is pumped away from the cold plates to a Coolant Distribution Unit (CDU), where the heat is dissipated to air or liquid depending on the type of CDU in place. Coolants can be either a water glycol mix or dielectric fluid.
Single Phase Immersion Cooling
The entire server including GPUs, CPUs amd DPUs is submerged in a thermally conductive, dielectric liquid coolant. The liquid coolant absorbs heat from the components and is then circulated through a heat exchanger to dissipate the heat. As the liquid coolant is never allowed to boil and turn into vapor, it is called single phase immersion cooling since the liquid remains in one state. This method provides excellent cooling efficiency and density, allowing for very high-density rack configurations.
Dual Phase Immersion Cooling
The entire server including GPUs, CPUs and DPUs is submerged in a thermally conductive, dielectric liquid coolant. Unlike single phase immersion cooling, the heat from the chips is allowed to cause the liquid coolant to boil and turn into vapor, which removes the heat from the electronics. The vapor rises, cools and condenses, allowing the liquid coolant to fall as droplets back into the tank. The continuous cycle of turning liquid into vapor and vapor into liquid is the reason why the process is called two phase immersion cooling. This approach offers exceptional cooling capacity and efficiency.
