AMD Stream 2.0 正式版发布

Edison

http://developer.amd.com/gpu/ATIStreamSDK/Pages/default.aspx

文档：
ATI Stream SDK v2.0 Documentation


Home > GPU Tools > ATI Stream Software Development Kit (SDK) v2.0 With OpenCL™ 1.0 Support > ATI Stream SDK v2.0 Documentation

Documents ATI Stream SDK Installation Notes (v2.0) [PDF 58.5KB] ATI Stream SDK Developer Release Notes (v2.0) [PDF 51.6KB] ATI Stream SDK Samples Release Notes (v2.0) [PDF 39.2KB] ATI Stream SDK Frequently Asked Questions (FAQ) (v2.0) [PDF 43.9KB] ATI Stream SDK Getting Started Guide (v2.0)[PDF 57.9KB] OpenCL™ 1.0 Specification (revision 48) [PDF 2.47MB] ATI Stream SDK OpenCL™ Performance and Optimization Notes (v2.0) [PDF 83.9KB] ATI Intermediate Language (IL) Specification (v2) [PDF 1.95MB] AMD Evergreen-Family ISA Microcode And Instructions (v1.0) [PDF 1.24MB] Knowledge Base Articles KB71 – Updating Your OpenCL™ Code To Work With The OpenCL™ ICD [HTML] KB83 – Common OpenCL™ ICD Error Messages [HTML] KB84 – Preview Feature: ATI Stream SDK v2.0 Support For OpenCL™ / OpenGL® Interoperability [HTML] KB85 – Preview Feature: ATI Stream SDK v2.0 Support For OpenCL™ / Microsoft® DirectX® 10 Interoperability [HTML] KB86 – Preview Feature: ATI Stream SDK v2.0 Support For Experimental OpenCL™ C++ Bindings [HTML] KB88 – Preview Feature: ATI Stream SDK v2.0 Support For Double-Precision Floating Point Basic Arithmetic In OpenCL™ C Kernels [HTML] KB19 – Running ATI Stream Applications Remotely [HTML] KB75 – Tips And Suggestions For Running SiSoftware Sandra 2010 OpenCL™ GPGPU Benchmarks With ATI Stream SDK v2.0 [HTML] OpenCL and the OpenCL logo are trademarks of Apple Inc. used by permission by Khronos.

你需要 AMD 注册开发者账号才能下载。

sagecao

感觉没什么用途，现在对所谓的通用运算一点性趣都没...........

哎

牛奶不老

相关应用需不需要安装这个？

Edison

Details and Recommendations


This preview feature is being offered to developers by AMD and does not constitute support for any official optional double-precision floating point extension as defined in the OpenCL™ 1.0 Specification. This feature may behave differently from what is documented in the specification.

Double-precision floating point addition, subtraction and multiplication operators are available for use in OpenCL™ C kernels.

Double-precision floating point built-in function support will be added in a future release.

Use of double-precision floating point arithmetic on GPUs other than ATI Radeon™ HD 5800 series GPUs and ATI Radeon™ HD 5900 series GPUs may result in unexpected behavior.

jocover

5770、5750不支持double precision

Edison

5770、5750不支持double precision
jocover 发表于 2009-12-22 08:57

RV830 不支持 DP FP。

不过 opencl 这里的主要问题是，amd 目前的这个 SDK 2.0 其实并未提供 DP 支持， 严格来说它只是提供了四则运算的 DP 支持，而 opencl 中其实有很多的函数，例如三角函数，这部分 SDK 2.0 就未能支持了。

正式的 DP 支持应该在 2010 年上半年，拖沓这么久的原因估计是 AMD 的 RV870 DP 单元有些地方没有完全遵循 opencl 1.0 的要求，需要一些软件上的动作。

lemonninja

据说R600/RV670被抛弃了呀。。。相比CUDA依然是不完善呀不完善。。。

ATI专业卡

AMD 把有限的人力投到游戏卡支持OPENCL 上。实在是无谓.

itany

站长居然不给个本地下载，喷、喷、喷！

enigmafeng

看不明白这是啥玩意，好像挺专业的。

yamhill

嗯，汇编2.0

也是杯具2.0