Investigation on Advanced Cold Plate Liquid Cooling Solution for Large Scale Application in Data Center

With the rapid development of computing capability and the increasingly stringent policies of energy conservation and carbon emission reduction from government regulation globally, efficient cooling technology has got more and more attention from industry. Meanwhile, high performance of CPU and GPU has been continuously requested to meet industrial demand, resulting in high power consumption and heat flux density of chips. Thermal Design Power (TDP) of Intel Xeon Scalable Processors has become higher and higher from generation to generation due to increasing core count, frequency, more caches and the complexity of multi-chip package(MCP). In the latest Intel server platform, TDP of CPU would be up to 350W and even higher in near future in terms of TDP and heat flux density, which exposes extreme challenge for system thermal solution design with traditional air cooling under current system form factor and thermal boundary condition. Nowadays, conventional air-cooling thermal solutions for high performance and high-power processor in server system have almost reached bottleneck even with maximum optimization on heatsink design and air flow improvement. And liquid cooling methodology, compared with traditional air-cooling technology, has more advantages, i.e. higher cooling capability, better energy efficiency and optimal PUE level etc.., technically appropriate to tackle industrial challenges and echo government's energy policy. This paper introduces the cold plate liquid cooling methodology developed by Kuaishou. COM with the latest generation of Intel Xeon Scalable Processors in detail, covering the design and optimization of cold plate, quick disconnector, tube and liquid leakage detection etc. Extensive tests have been performed with different design of experiment (DOE) parameters, studying the impact on the server's cooling capability from temperature of supply fluid, flow rate, lab ambient temperature and fan speed etc. The author also proposes the direction to decrease the total cost of ownership(TCO) of cold plate liquid solution and provides the total liquid cooling solution and application for ecosystem and industry's reference.