CCF: EAGER: DeepGreen: Modeling and Boosting Accelerated Computing on Liquid Immersion Cooled HPC Systems

功率和散热问题已成为限制高性能计算系统和数据中心的性能、可扩展性和可负担性的关键设计限制因素。随着服务器越来越密集并采用加速器，应对这一挑战是当务之急。液体浸没冷却是一种潜在的解决方案，因为它比传统的空气冷却具有更好的导热性能和更低的功率需求。该项目探索沉浸式冷却，以提高采用多核和多核处理器和加速器的高密度服务器的能效、计算能力和可靠性。它吸引研究生和本科生从事研究，并激励和培养技术和工程领域的下一代劳动力。该项目倡导绿色计算，旨在降低服务器和数据中心的能源成本。该项目与一家液体浸泡冷却供应商密切合作，将有助于技术创新和转让。该项目的目标是高密度服务器，包括图形处理单元(GPU)、张量处理单元(TPU)、现场可编程门阵列(FGA)和专用集成电路(ASIC)等加速器。它探索了高级机器学习，以自动从多物理、不同数据中学习浸泡冷却对性能、功率和热量的影响，并设计模型辅助管理方案，以满足包括能效、性能和热点消除在内的各种优化目标。该项目的完成将在多个方面推进最先进的加速计算和数据中心冷却，包括定量和全面地了解浸泡冷却的好处，更准确的性能、功率和加速系统的热量建模，以及智能管理软件。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Power and thermal issues have become critical design constraints that limit performance, scalability, and affordability for high performance computing systems and datacenters. It is urgent to address this challenge as servers are increasingly dense and employing accelerators. Liquid immersion cooling is a potential solution, due to its significantly better thermal conduction and lower power demand than traditional air cooling. This project explores immersion cooling to boost energy efficiency, computing capacity, and reliability of dense servers with multicore and manycore processors and accelerators. It engages both graduate and undergraduate students in research, and inspires and fosters the next generation workforce in technology & engineering. This project promotes green computing, which seeks to reduce energy cost of servers and datacenters. The project features close collaboration with a liquid immersion cooling vendor and will help technology innovation and transfer. This project targets dense servers comprising accelerators such as graphics processing units (GPUs), tensor processing units (TPUs), field programmable gate arrays (FPGAs), and application-specific integrated circuits (ASICs). It explores advanced machine learning to automatically learn from multi-physics, diverse data the impact of immersion cooling on performance, power, and thermal, and design model-assisted management schemes to meet various optimization objectives including energy efficiency, performance, and hotspot elimination. Completion of this project will advance the state-of-the-art accelerated computing and datacenter cooling in multiple aspects, including quantitative and comprehensive understanding of the benefits of immersion cooling, more accurate performance, power, and thermal modeling for accelerated systems, and intelligent management software.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.