Recent advancements in computational fluid dynamics(CFD) driven by the heterogeneous computing techniques and high-fidelity numerical methods have significantly increased the demand for efficient input/output(I/O) operations. In GPU-accelerated CFD, redundant data copies and excessive CPU overhead have become prominent challenges for efficient IO due to increasing complexities in data transfers between memory and storage. In this work, we propose a GPU native I/O framework(named as GPUDirectIO) for high-performance CFD by redesigning the Data Mapping Layer(DML) and data structures of the CFD General Notation System(CGNS), which is a widely used file format for complex CFD applications. This GPU-centric system enables direct data transfers between NVMe storage and GPU memory via GPU Direct Storage(GDS), effectively streamlining heterogeneous CFD I/O workflows. To further improve I/O throughput, we develop a CGNS-based distributed data management strategy that leverages an NVMe storage array to fully utilize the GPU bandwidth. We compare the performance of the proposed GPUDirectIO with existing CPU-mediated I/O approaches with different number of threads using CFD datasets where the maximum number of computa tional grid points reaches 1.6 billion. The results demonstrate the superiority of GPUDirectIO. Single-threaded GPUDirectIO achieves read and write rates 2.95× and 3.49× those of CPU mediated I/O, respectively, and reduces transmission latency by approximately 59%. When applied to distributed storage systems, multi-threaded GPUDirectIO shows read and write rates 3.23× and 4.68× those of CPU-mediated parallel I/O, respectively, with transfer latency reduced by about 39%. GPUDirectIO has also demonstrated excellent parallel efficiency and speedup ratios in both strong and weak scalability tests.

Summary The development of computational fluid dynamics (CFD) highly depends on high‐performance computers. Computer hardware has evolved rapidly, yet scalable CFD parallel software remains scarce. In this article, we design a highly scalable CFD parallel paradigm for both homogeneous and heterogeneous supercomputers. The paradigm achieves the separation of communication and computation and automatically adapts to various solvers and hardware environments, thus reducing programming difficulties and increasing automatic parallelization. Meanwhile, the number of communications is greatly reduced and the scalability of the program is improved through implementing centralized communication and two‐level partitioning techniques. Complex flow problems for real aircraft were then computed on different hardware platforms with a grid size of ten billion. The homogeneous computer hardware includes Intel Xeon Gold 6258R and Phytium 2000+ processors, and the heterogeneous computer platforms include NVIDIA Tesla V100 and SW26010 processors. High parallel efficiency was obtained on all computer platforms, verifying that the paradigm has good automatic parallelization, scalability, and stability. The paradigm in this article has an important reference value for CFD massively parallel computing and can promote the development and application of CFD technology.

Solving partial differential equations of complex physical systems is a computationally expensive task, especially in Computational Fluid Dynamics(CFD). This drives the application of deep learning methods in solving physical systems. There exist a few deep learning models that are very successful in predicting flow fields of complex physical models, yet most of these still exhibit large errors compared to simulation. Here we introduce AMGNET, a multi-scale graph neural network model based on Encoder-Process-Decoder structure for flow field prediction. Our model employs message passing of graph neural networks at different mesh graph scales. Our method has significantly lower prediction errors than the GCN baseline on several complex fluid prediction tasks, such as airfoil flow and cylinder flow. Our results show that multi-scale representation learning at the graph level is more effective in improving the prediction accuracy of flow field.