优化图形流水线OptimisingtheGraphicsPipeline优化图形.pdfVIP

优化图形流水线 优化图形流水线 Optimising the Graphics Pipeline Optimising the Graphics Pipeline Koji Ashida Hi. My name is Koji Ashida. Im part of the developer technology group at NVIDIA and we are the guys who tend to assist game developers in creating new effects, optimizing their graphics engine and taking care of bugs. Today, Im going to talk to you about optimizing the graphics pipeline. 概述 CPU GPU application driver 潜在的瓶颈部位 瓶颈（ bottleneck ）制约着总的数据吞吐率 总的来说，每个应用程序甚至每一帧都会有不同的瓶颈部位 对于流水线（ pipeline ）架构来说，要获得好的性能取决于发现和 排除瓶颈的影响 Were going to go through all the stages of the GPU plus talk a little bit about what happens on the CPU side that can bottleneck a graphics application. As we all know, in order to render a scene, the application which is running on the CPU must send data and instructions across to the device. It communicates with the device through the driver. Then, once the device has the data, it processes the data using the graphics chip itself, and finally writes it out to the frame buffer. Because this whole process is a single pipeline from the CPU to the last stage of the GPU, any one of those stages can be a potential bottleneck. The good thing about a pipeline is that its very efficient and, in particular, its very efficient at rendering graphics because it can parallelize a lot of operations. The bad thing about a pipeline is that, once you do have a bottleneck, then the whole pipeline is running at that speed so you really want to basically level off all your stages in the pipeline such that they have equal workloads. Hopefully, by the end of this presentation, youll have a very good idea of how to either reduce the workload on a certain stage or at least increase the wor