基于光线追踪机制的三维集成图形处理器体系结构研究

项目名称： 基于光线追踪机制的三维集成图形处理器体系结构研究

项目编号： No.61272085

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 邓仰东

作者单位： 清华大学

项目金额： 80万元

中文摘要： 图形显示是人机交互的核心手段，目前几乎所有个人计算机、平板电脑和智能手机都装备图形处理单元(Graphics Processing Unit, GPU)来实现显示功能。光栅化和光线追踪是实现图形显示的两类主要算法，前者被主流商业GPU所广泛使用，而后者允许高度逼真的光影效果、并在高度复杂场景下具有较低的算法复杂度。由于人类对于显示质量的要求日益提高和显示场景复杂度快速提升，光线追踪算法在满足新一代图形应用方面具有巨大潜力，有望成为下一代主流图形显示方法。本课题在深入分析光线追踪算法的基础上，探索基于光线追踪算法的众核图形处理器体系结构(微架构，microarchitecture)。通过发展具备良好扩展性的众核体系结构和兼顾不同计算模式的组织与调度方式、并利用三维集成电路消除传统GPU体系结构的性能瓶颈，本课题的设计目标为在1280x1024分辨率下实现接近实时性能(～15帧/秒)的显示速率

中文关键词： 光线追踪；图形处理器；空间数据结构；3-D；近眼光场显示

英文摘要： Graphics rendering is the essential method of human-computer interface. Now almost all computers, tablets and smartphones are equipped with one or more graphics processing units (GPUs) for graphics processing. Currently, rasterization and ray-tracing are the 2 dominant rendering algorithms. Although the former is widely deployed in commercial GPUs, the latter promises realistic rendering effects and a lower complexity on highly complex scenes. Due to the ever-growing demand for high graphics quality and scene complexity, the advantages of ray-tracing is likely to make it become the mainstream rendering method. Starting from a detailed performance profiling of the ray-tracing algorithm on typical scenes, this project explores a many-core microarchitecture for ray-tracing. The cores on the proposed microarchitecture are organized in a scalable manner and designed to simultaneously support diverse computing patterns. The microarchitecture also take advantage of the 3-D IC technology to remove the performance bottlenecks in traditional GPUs. The objective of the microarchitecture design is to deliver a frame rate of around 15 fps at a resolution of 1280x1024.

英文关键词： Ray tracing；GPU；Spatial accleration structure；3-D；Near-eye light-field display