发表机构
ETH Zurich（苏黎世联邦理工学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文质疑科学计算中FP64高精度算术的必要性，提出基于信噪比的准则评估安全降精度，并设计自动化数据流工作流，在GPU上对天气气候应用实现最高1.8倍加速且不损物理保真度。
AI 中文摘要
尽管FP64（binary64）仍然是科学计算中实数的标准表示，但硬件趋势日益倾向于针对AI工作负载优化的低精度格式。这种转变优先考虑吞吐量和能效，而非全精度能力。我们质疑在受偏微分方程（PDE）控制的模拟中高精度算术的必要性，在这些模拟中，空间和时间离散化误差引入了物理噪声基底，掩盖了FP64格式的低位比特，使其变得无关紧要。我们提出了一种基于信噪比（SNR）相对于空间和时间细化的准则，以评估安全降精度对科学计算的影响。最后，我们提出了一种自动化的、以数据流为中心的工作流，用于在复杂科学应用的混合精度子图中进行有针对性的降精度。对天气和气候应用的大规模评估表明，在GPU上进行策略性精度降低可实现高达1.8倍的加速，且不损害物理保真度。
英文摘要
While FP64 (binary64) remains the standard representation of real numbers in scientific computing, hardware trends increasingly favor low-precision formats optimized for AI workloads. This shift prioritizes throughput and energy efficiency over full-precision capabilities. We challenge the necessity of high-precision arithmetic in PDE-governed simulations, where spatial and temporal discretization errors introduce a physical noise floor that masks the lower-order bits of the FP64 format, rendering them irrelevant. We propose a criterion based on the Signal-to-Noise Ratio (SNR) with respect to spatial and temporal refinements to evaluate the impact of safe precision reduction on scientific computations. Finally, we present an automated dataflow-centric workflow for targeted lowering within mixed-precision subgraphs in complex scientific applications. Large-scale evaluations of weather and climate applications show that strategic precision reduction on GPUs achieves up to 1.8x speedup without compromising physical fidelity.
CommentsAccepted at SC '26: The International Conference for High Performance Computing, Networking, Storage, and Analysis, November 15-20, 2026, Chicago, IL, USA. 15 pages, 10 figures, 4 tables
