Journal of Computer Applications

    Next Articles

State-of-The-Art Survey on Hardware Acceleration for Ray Tracing

  

  • Received:2024-05-09 Revised:2024-10-02 Accepted:2024-10-12 Online:2024-11-22 Published:2024-11-22
  • Contact: Zong-Hui ZongLi

光线追踪硬件加速方案综述

张大权1,董家瑞1,雷洋1,李世康1,石响宇1,李宗辉1,邓仰东2,吴为民1   

  1. 1. 北京交通大学计算机与信息技术学院
    2. 清华大学
  • 通讯作者: 李宗辉

Abstract: Nowadays, real-time 3D graphics rendering is undergoing technological innovation, and one of the most important trends is the surging requirements for real-time ray tracing. But ray tracing is still "expensive" in terms of computing, and traditional hardware cannot support such computing power. New graphics processing units (GPUs) must balance the cost among performance, power consumption, and higher complexity scenarios, so hardware acceleration is the key core of ray tracing. First, this paper introduces the theoretical basis of ray tracing, and based on the two most dominant space accelerated data structures (KD-Tree and BVH-Tree), this paper investigates primitive partitioning, construction methods, optimization methods, and traversal acceleration to reveal the potential for hardware acceleration. Second, this paper investigates the dedicated acceleration hardware corresponding to three perspectives, namely fixed-function design, scheduling and data management to reduce memory bandwidth, and hardware architecture design. Furthermore, this paper investigates industry solutions for real-time ray tracing as well as industry trends of hardware acceleration in the coming years. Finally, this paper discusses and concludes the accelerated hardware status and deficiency to point out the future research directions for both academia and industry. This paper is well suitable for researchers and developers including beginners and specialists who work on ray tracing.

Key words: Ray Tracing, Hardware Acceleration, KD-tree, BVH, GPUs

摘要: 当前,实时三维图形渲染领域发生着技术变革,实时光线追踪技术的应用在激增;但就计算而言,光线追踪成本依旧“昂贵”,传统硬件无法支持这样的算力。新的图形处理单元(GPU)必须在性能、功耗和高复杂度场景之间获取平衡,硬件加速技术因此成为实时光线追踪的核心。首先,本文介绍了光线追踪的理论基础,在基于目前最主流的两种硬件加速数据结构(KD树和BVH树)的基础上,分别从基元分割、构造方法、优化方法和遍历加速的角度进行了调研,发掘可用于硬件加速的潜力;然后,从固定函数设计、硬件架构设计、以减少内存带宽为目标的调度和数据管理三个角度,对各个阶段所开发的专用硬件进行了综述;随后,面向产业界调研了主流光线追踪图形处理器的业界解决方案以及未来发展趋势;最后,总结并讨论了光线追踪硬件加速方案的现状与不足,为学术界和产业界提供潜在的性能优化方向。本文适合各类从事光线追踪研究和研发的人员包括初学者和资深研究者。

关键词: 关键词: 光线追踪, 硬件加速, KD树, BVH树, GPUs

CLC Number: