湍流燃烧算例在多种CPU平台的模拟性能对比

Computational Performance Comparison of CPU Platforms in Turbulent Combustion Simulation

新型燃烧室研发需要高精度数值模拟及大规模并行计算。几何高保真加上高精度的湍流模型、湍流燃烧模型和化学反应机理等,必需从各方面提升大规模并行计算效率。目前超算平台的浮点运算能力评测并不能完全代表湍流燃烧模拟的实际消耗,因此列举了各计算平台的Linpack测试分数进行参考,采用典型湍流火焰数值模拟对常用的CPU超算平台进行测试,检验并行计算效率,分析结果。湍流燃烧模拟采用自研AECSC软件,结构化网格,对比了Intel、AMD和国产CPU平台。结果表明,超算平台浮点数运算能力和湍流火焰模拟效率不一致,对同一算例的计算时间存在差异;国产超算平台表现优于参考的浮点算力;不同的求解过程有不同的加速比。

The development of new combustors requires high-precision numerical simulation and large-scale parallel calculation.It is necessary to improve the computation efficiency to combine high fidelity geometric,high precision turbulence models,turbulent combustion models and chemical reaction mechanisms.It is not entirely accurate to judge combustion simulation consumption by floating point arithmetic capability.Thus this paper firstly enumerated the Linpack test scores of each computing platform.Then a typical turbulent flame numerical simulation is used to test the super calculate platform to verify the parallel computing efficiency.The simulation was based on the self-developed software AECSC with structured grids.The platform of Intel,AMD and Domestic CPU was compared.The results show that the floating point arithmetic capability of the super calculate platform is not exactly consistent with the turbulent flame simulation efficiency,and the calculation time of the same example on different platforms is different.The performance of domestic super calculate platforms is better than the floating point arithmetic reference.And different solution procedures have different acceleration ratios.