Streamline simulation in groundwater flow modeling is a time‐consuming process when a large number of streamlines are analyzed. We develop a parallelization method on graphics processing units (GPUs) for the semi‐analytical particle tracking algorithm developed by Pollock (1988). Compute Unified Device Architecture was used to implement the parallel method. Forward and backward tracking of a streamline is handled by an individual thread. A GPU includes a grid of blocks where a block handles 32 threads. We use multi‐GPUs to accelerate streamline tracking in a flow model with millions of particles. The method was examined to simulate streamlines for identifying three‐dimensional (3D) flow systems in a Tóthian basin. The speedup exceeds 1000 when 8 NVIDIA GPUs are used to simulate 5 million or more streamlines.

中文翻译：

---

在GPU上进行地下水流建模时加快流线跟踪

当分析大量流线时，地下水流模型中的流线模拟是一个耗时的过程。我们针对Pollock（1988）开发的半解析粒子跟踪算法，在图形处理单元（GPU）上开发了并行化方法。计算统一设备体系结构用于实现并行方法。流线的向前和向后跟踪由单个线程处理。GPU包括一块块的网格，其中一块可处理32个线程。我们使用多GPU来加速具有数百万个粒子的流模型中的流线跟踪。检查了该方法以模拟流线，以识别Tóthian盆地的三维（3D）流动系统。当使用8个NVIDIA GPU模拟500万或更多流水线时，加速超过1000。