Copper or Aluminum , which is better for liquid cooling solution
With the rapid development of artificial intelligence technology, especially in fields such as deep learning and large-scale language models, the demand for computing power has significantly increased. Today's AI models, such as GPT-4o, have tens or even billions of parameters and require enormous computing resources for training. Training these models requires a large number of GPU or TPU clusters, which generate a significant amount of heat when running at full load. In addition, in order to provide real-time response in applications, many AI systems require continuous operation. These systems are usually deployed in the data center or edge computing devices, which also face high power consumption and cooling challenges.
With the advancement of chip technology and the rapid growth of server computing power, building high-density, high-energy consumption large data centers has become a necessary choice to balance computing power and environmental regulations. The refrigeration system is one of the important infrastructure in data centers. In high-density data center operation, traditional air cooling faces problems of insufficient heat dissipation and serious energy consumption. Liquid cooling technology has become the optimal solution to reduce PUE in data centers, with more economic advantages at 15kW/cabinet and above.
Liquid cooling plate technology is a thermal solution that indirectly transfers the heat of components to a cooling liquid enclosed in a circulating pipeline through a cold plate (a closed cavity composed of high thermal conductivity metals such as copper and aluminum), and then uses the cooling liquid to take away the heat.
Liquid Cold plate is the earliest adopted liquid cooling method, with high maturity and relatively low price. According to research data, cold plate liquid cooling accounts for 90% of the market share in China. Cold plate liquid cooling is achieved by tightly fixing the cold plate to the heating element, transferring the heat from the heating element to the cooling liquid in the cold plate. It is simple, rough, but effective. The penetration rate of liquid cooling technology in data centers is expected to be around 5% to 8% in 2022, with air cooling still holding over 90% of the market share.
The thermal conductivity of copper is about 400 W/mK, and the thermal conductivity of aluminum is about 235 W/mK. The thermal conductivity of copper is much higher than that of aluminum. Therefore, copper cold plates can theoretically transfer the heat generated by servers more quickly to the coolant, thereby achieving more efficient heat dissipation. Although the thermal conductivity of aluminum is not as good as copper, its thermal conductivity is relatively high, which is sufficient to meet the heat dissipation needs of most liquid cooled servers.
The density of copper is relatively high, about 8.96 g/cm ³, which makes the copper cold plate relatively heavy. This may pose certain challenges to the structural design and installation of the server. Aluminum has a lower density of about 2.70 g/cm ³, which is much lighter than copper, so aluminum cold plates have a significant advantage in weight. The low density of aluminum makes aluminum cold plates lighter. This is not only beneficial for reducing the overall weight of the server, but may also improve the structural strength of the server to a certain extent. In addition, aluminum material is lighter, which is beneficial for reducing the overall weight of servers and lowering transportation and installation costs.
Copper and aluminum cold plates have their own advantages and disadvantages in the use of liquid cooled servers. In situations where thermal requirements are high and cost is not the main consideration, copper cold plates may be more suitable; In the pursuit of cost-effectiveness and lightweight, aluminum cold plates may have more advantages. The specific selection needs to be comprehensively considered based on the requirements and limitations of the specific application scenario. If we can have a detailed understanding of the specific situations such as heat load, budget, weight restrictions, etc. in the application scenario, it can help us make more accurate choices.
