Abstract
In recent years, rapid growth is seen in computer and server processors in terms of thermal design power (TDP) envelope. This is mainly due to increase in processor core count, increase in package thermal resistance, challenges in multi-chip integration and maintaining generational performance CAGR. At the same time, several other platform level components such as PCIe cards, graphics cards, SSDs and high power DIMMs are being added in the same chassis which increases the server level power density. To mitigate cooling challenges of high TDP processors, mainly two cooling technologies are deployed: Liquid cooling and advanced air cooling. To deploy liquid cooling technology for servers in data centers, huge initial capital investment is needed. Hence advanced air-cooling thermal solutions are being sought that can be used to cool higher TDP processors as well as high power non-CPU components using same server level airflow boundary conditions.
Current air-cooling solutions like heat pipe heat sinks, vapor chamber heat sinks are limited by the heat transfer area, heat carrying capacity and would need significantly more area to cool higher TDP than they could handle. Passive two-phase thermosiphon (gravity dependent) heat sinks may provide intermediate level cooling between traditional air-cooled heat pipe heat sinks and liquid cooling with higher reliability, lower weight and lower cost of maintenance. This paper illustrates the experimental results of a 2U thermosiphon heat sink used in Intel reference 2U, 2 node system and compare thermal performance using traditional heat sinks solutions. The objective of this study was to showcase the increased cooling capability of the CPU by at least 20% over traditional heat sinks while maintaining cooling capability of high-power non-CPU components such as Intel’s DIMMs. This paper will also describe the methodology that will be used for DIMMs serviceability without removing CPU thermal solution, which is critical requirement from data center use perspective.