17.2 What is Graphics Hardware Graphics hardware describes the hardware components necessary to quickly render 3D objects as pixels on your computer's screen using specialized rasterization-based (and in some cases, ray-tracer-based) hardware architectures. The use of the term graphics hardware is meant to elicit a sense of the physical components necessary for performing a range of graphics computations. In other words, the hardware is the set of chipsets, transistors, buses, processors, and computing cores found on current video cards. As you will learn in this chapter, and eventually experience yourself, current graphics hardware is very good at processing. Using Graphics Hardware descriptions of 3D objects and transforming those representations into the colored pixels that fll your monitor. Real-Time Graphics: By real-time graphics, we generally mean that the graphics-related computations are being carried out fast enough that the results can be viewed immediately. Being able to conduct operations at 60Hz or higher is considered real time. Once the time to refresh the display (frame rate) drops below 15Hz, the speed is considered more interactive than it is real-time, but this distinction is not critical. Because the computations need to be fast, the equations used to render the graphics are often approximations to what could be done if more time were available. Graphics hardware has certainly changed very rapidly over the last decade. Newer graphics hardware provides more parallel processing capabilities, as well as better support for specialized rendering. One explanation for the fast pace is the video game industry and its economic momentum. Essentially what this means is that each new graphics card provides better performance and processing capabilities. As a result, video games appear more visually realistic. The processors on graphics hardware, often called GPUs, or Graphics Processing Units, are highly parallel and afford thousands of concurrent threads of execution. The hardware is designed for throughput which allows larger numbers of pixels and vertices to be processed in shorter amounts of time. All of this parallelism is good for graphics algorithms, but other work has benefited from the parallel hardware. In addition to video games, GPUs are used to accelerate physics computations, develop real-time ray tracing codes, solve Navier-Stokes related equations for fluid flow simulations, and develop faster codes for understanding the climate. Several APIs and SDKs have been developed that afford more direct general purpose computation, such as OpenCL and NVIDIA's CUDA. Hardware accelerated ray tracing APIs also exist to accelerate ray-object intersection. Similarly, the standard APIs that are used to program the graphics components of video games, such as OpenGL and DirectX, also allow mechanisms to leverage the graphics hardware's parallel capabilities. Many of these APIs change as new hardware is developed to support more sophisticated computations. Graphics hardware is programmable. As a developer, you have control over much of the computations associated with processing geometry, vertices, and the fragments that eventually become pixels. Recent hardware changes as well as Fragment: Fragment is a term that describes the information associated with a pixel prior to being processed in the finnal stages of the graphics pipeline. This definition includes much of the data that might be used to calculate the color of the pixel, such as the pixel's scene depth, texture coordinates, or stencil information. Ongoing updates to the APIs, such as OpenGL or DirectX, support a completely programmable pipeline. These changes afford developers creative license to exploit the computation available on GPUs. Prior to this, fixed-function rasterization pipelines forced the computation to a specific style of vertex transformations, lighting, and fragment processing. The fxed functionality of the pipeline ensured that basic coloring, lighting, and texturing could occur very quickly. Whether it is a programmable interface, or fixed-function computation, the basic computations of the rasterization pipeline are similar, and follow the illustration in Figure 17.1. In the rasterization pipeline, vertices are transformed from local space to global space, and eventually into screen coordinates, after being transformed by the viewing and projection transformation matrices. The set of screen coordinates associated with a geometry's vertices are rasterized into fragments. The final stages of the pipeline process the fragments into pixels and can apply.