| 南统超,施睿,于春水.基于国产CPU/GPU平台的水动力模型异构并行计算[J].水利学报,2025,56(6):771-779 |
| 基于国产CPU/GPU平台的水动力模型异构并行计算 |
| Hydrodynamic modeling via heterogeneous parallel computing on domestic CPU/GPU platform |
| 投稿时间：2024-06-01 |
| DOI：10.13243/j.cnki.slxb.20240326 |
| 中文关键词: 偏微分方程求解 水动力模型 并行计算 异构计算 国产处理器 洪水模拟 |
| 英文关键词: partial differential equations hydrodynamic models parallel computing heterogeneous computing domestic processors flood simulation |
| 基金项目:国家自然科学基金面上项目(42472307);中央高校基本科研业务费项目(B240201058);光合基金A类项目(202202049048) |
|
| 摘要点击次数: 46 |
| 全文下载次数: 25 |
| 中文摘要: |
| 水动力模型具有物理意义明确、可靠性高等优点，但其计算量大、算力要求高。鉴于先进计算芯片进口遇阻，国外CUDA软件平台也被禁用，有必要基于国产CPU/GPU产业现状发展“多节点-多GPU”异构并行计算方案。本研究基于国产硬件平台研发“多节点-多GPU”大规模分布式异构并行计算水动力模型，用于流域尺度二维浅水方程高效并行求解。通过三个算例检验了该模型的准确性、并行效率以及对亿级大模型的模拟能力。结果表明，该模型计算准确，并行加速效果显著，可模拟溃坝、大尺度复杂地形等条件下水动力过程，实现了对亿级网格洪水过程分钟级模拟。有望为我国洪水预报预警提供自主可靠的技术支撑，为流域大模型、区域地下水模型等其他大型水动力模型的异构并行计算提供借鉴。 |
| 英文摘要: |
| Hydrodynamic models have advantages such as clear physical meaning，high precision and reliability，but they also have higher computational requirements and more need for computing power.Given the restless international environment，it is necessary to develop a “multi-node，multi-GPU” heterogeneous parallel computing scheme based on domestically-produced CPU/GPUs.This study developed a “multi-node，multi-GPU” large-scale distributed hydrodynamic model on heterogeneous parallel computing platform based on domestically-produced devices，for fast solving the two-dimensional shallow water equations at the watershed scale.Through three examples of increasing complexity，the accuracy，parallel efficiency，and simulation capability of large-scale models with hundreds of millions of grids were investigated.The results show that the proposed model can accurately simulate the hydrodynamic process under conditions such as dam break and large-scale complex terrain;the parallel acceleration effect is significant，achieving minutes-level fast simulation of flood processes on hundreds of millions of grids.It is expected to provide independent and reliable technical support for China’s flood forecasting and early warning，and provide reference for parallel computing of other hydrodynamic models such as large-scale basin models and regional groundwater models. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |