[CUDAGPU入门介绍.pptVIP

--An introduction to beginners GPU CUDA Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. 内容 1 GPU高性能计算 2 CUDA架构 3 CUDA环境搭建 4 CUDA简单例子 Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. 1. GPU高性能计算 GPU: 从图形处理到通用计算 GPU: Graphic Processing Unit GPU为高度并行的实时3D渲染计算而设计, 高GFLOPS , 高带宽 3D渲染技术及3D API的发展, 促进GPU向通用计算处理器发展 NVIDIA GPU为通用计算专门优化设计, 于2007年推出CUDA Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. 1. GPU高性能计算 GPU VS CPU: 计算能力 8x double precision ECC L1, L2 Caches 1 TF Single Precision 4GB Memory Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. 1. GPU高性能计算 GPU VS CPU: 硬件体系 目标: 最小的指令延迟 巨大的缓存 复杂的控制逻辑 目标: 最大吞吐量 更多寄存器、高带宽 通过线程轮换隐藏延迟 多个线程共享控制逻辑 Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. 1. GPU高性能计算 GPU应用领域 用同一计算方法对很大量的数据进行并行计算 数据相关度低 计算密度大，计算所花的时间比数据存储的时间大得多 Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. Evaluation only. Created with Aspose.Slides for .NET 3.5 Client Profile . Copyright 2004-2011 Aspose Pty Ltd. 1. GPU高性能计算 GPU关键概念 SM: Stream Multiprocessor SP: Stream Processor 每个SM包含8个SP, 由SM取指, 解码, 发射到各个SP, GPU可看作是一组SM SMEM: shared memory, GPU片内每个SM所占有的高速存储器 Global me