GPU通用计算及其在计算智能领域的应用

摘要

The GPU enjoys the characteristics of high parallelism , low energy consumption and cheap price .Com-pared with the traditional CPU platform , it is especially suitable for tasks with high data parallelism .GPU compu-ting has come into the mainstream of high performance computation ( HPC) due to the emerging of development platforms like CUDA and OpenCL .The GPU's enormous computational power greatly promotes computational intel-ligence .A great success has been achieved in the fields such as deep learning and swarm intelligence optimization , and several breakthroughs have been seen in image , and speech recognition because of GPU .Though suffering some drawbacks, GPUs provide common people and small institutions with enormous computing power .This has changed the set-up of scientific computing and programming model because it could only be provided by expensive super -computers.To help researchers in the field of computational intelligence better utilize GPUs , a detailed survey of GPGPU is given in this paper. First, the characteristics and advantages of GPUs against CPUs are presented .Then we briefly review the development of GPU hardware followed by a survey of the evolution of development tools for GPGPU;special attention is drawn to two major platforms , CUDA and OpenCL .We end this paper with our per-spectives of the challenges and trends of GPGPU .We point out that embedding and cluster are two major trends for GPGPU and as both academia and industry continue to see increasing progress in artificial intelligence , the GPU will be more widely used in more domains .%在日趋复杂的图形处理任务的推动下，GPU已经演化成为具有众多计算核心、计算能力强大的通用计算设备，并被越来越多地应用于图形处理之外的计算领域。 GPU具有高并行、低能耗和低成本的特点，在数据并行度高的计算任务中，相比与传统的CPU平台有着显著的优势。随着GPU体系结构的不断演进以及开发平台的逐步完善，GPU已经进入到高性能计算的主流行列。 GPU通用计算的普及，使个人和小型机构能有机会获得以往昂贵的大型、超级计算机才能提供的计算能力，并一定程度上改变了科学计算领域的格局和编程开发模式。 GPU提供的强大计算能力极大地推动了计算智能的发展，并且已经在深度学习和群体智能优化方法等子领域获得了巨大的成功，更是在图像、语音等领域取得了突破性的进展。随着人工智能技术和方法的不断进步，GPU将在更多的领域获得更加广泛的应用。