Direct numerical simulations (DNS) of turbulent flows have increasing importance because they not only provide fundamental understanding of turbulent flows but also complement and extend experimental results. DNS of high Reynolds numbers, however, require huge computing cost so high-performance computing has been strongly pursued. In this study, we examine the feasibility of cost-efficient DNS on Intel Xeon Phi many-core processors that are currently adopted by 10% of the 100 largest supercomputers in the world as listed in the Top500 site. For this purpose, we port and optimize our in-house turbulent flow solver named as DNS-TBL (direct numerical simulation-turbulent boundary layer) on Xeon Phi Knights Landing (KNL) many-core processors and conduct benchmark tests on KNL and conventional multicore processors. The key architectural features of KNL processors and strategies to exploit them for performance enhancement are discussed. The optimized code is validated by conducting numerical simulations of zero-pressure gradient turbulent boundary layers at high Reynolds numbers and by comparing simulated turbulent statistics to those reported in previous studies. With the details of optimization strategies and validation processes, this work can serve as a practical guideline for acceleration of large-scale and precise DNS with many-core computing.

湍流的直接数值模拟（DNS）越来越重要，因为它们不仅提供了对湍流的基本了解，而且补充并扩展了实验结果。但是，高雷诺数的DNS要求巨大的计算成本，因此一直强烈追求高性能计算。在这项研究中，我们研究了在英特尔至强融核多核处理器上采用具有成本效益的DNS的可行性，目前，全球500强超级计算机中有100％采用了英特尔至强融核多核处理器。为此，我们在至强披披骑士登陆（KNL）多核处理器上移植并优化了内部湍流求解器，称为DNS-TBL（直接数值模拟-湍流边界层），并在KNL和常规多核处理器上进行了基准测试处理器。讨论了KNL处理器的关键体系结构特征以及利用它们来提高性能的策略。通过对高雷诺数下的零压力梯度湍流边界层进行数值模拟，并将模拟的湍流统计数据与以前的研究结果进行比较，可以验证优化的代码。借助优化策略和验证过程的详细信息，这项工作可以作为使用多核计算加速大规模和精确DNS的实用指南。