CAREER: A systems approach to real-time graphics on single-chip highly-programmable hardware

william R. Mark，德克萨斯大学奥斯汀分校，该项目的目标是产生更好的交互式3D图形，即合成与现实世界无法区分的图像，并以足够快的速度生成这些图像，以便用户可以与计算机生成的世界进行交互。紧急和军事训练、电子商务、教育和娱乐等应用都需要逼真的计算机生成图像。虽然实时3D图形技术在过去的二十年里有了巨大的进步，但从今天的系统中获得的图像仍然不是完全真实的，而且由于当今系统中使用的技术的基本限制，现实主义的目标实际上是无法实现的。这项研究正在探索一种新的、基于软件和硬件的系统设计，以克服当前的限制，实现逼真的实时图形。新系统是围绕光线追踪可见性算法组织的，它不会受到使用z缓冲区所施加的限制。光线追踪传统上被认为对实时系统是不切实际的，因为以前的方法无法有效地支持大多数应用程序所需的动态和可变形对象。新系统通过一种新颖的光线追踪方法克服了这一限制，该方法集成了以前分离的系统的两个部分：场景管理和可见性计算。反过来，这种紧密集成要求将当前系统的两部分硬件体系结构（中央处理单元(CPU)+图形）替换为单个统一的硬件体系结构。这种新的硬件架构结合了当今图形处理器的并行性和当今cpu的灵活编程模型。这种新的硬件架构也有潜力为3D图形以外的各种应用提供更好的性能。结合这项研究，研究者正在培训学生和其他研究人员了解如何设计和分析这种新型系统。

AbstractCAREER: A systems approach to real-time graphics on single-chip highly-programmable hardwareWilliam R. Mark, University of Texas at Austin The goal of this project is to produce better interactive 3D graphics -- that is, to synthesize images that are indistinguishable from the real world and to generate these images rapidly enough that users can interact with the computer-generated world. Realistic computer-generated images are needed by applications such as emergency and military training, e-commerce, education, and entertainment. Although real-time 3D graphics technology has improved enormously over the past twenty years, the images from today's systems are still not fully realistic, and the goal of realism is in fact unattainable with current system designs due to fundamental limitations of thetechnology used in today's systems.This research is exploring a new, ground-up system design for both software and hardware that overcomes current limitations to enable realistic real-time graphics. The new system is organized around the ray tracing visibility algorithm, which does not suffer from the limitations imposed by using the Z-buffer. Ray tracing has traditionally been considered impractical for real-time systems because previous approaches have been unable to efficiently support dynamic and deformable objects that are needed by most applications. The new system overcomes this restriction with a novel approach to ray tracing that integrates the two parts of the system that were previously separated: scene management and visibility computations. In turn, this tight integration requires that the two-part hardware architecture of today's systems (the central processing unit (CPU)+graphics) be replaced with a single, unified hardware architecture. This new hardware architecture combines the parallelism of today's graphics processors with the flexible programming model of today's CPUs. This new hardware architecture also has the potential to provide better performance for a wide variety of applications beyond 3D graphics. In conjunction with this research, the investigator is training students and other researchers to understand how to design and analyze this new type of system.