ATI Stream 与 NVidia/CUDA 的比较
为了使这个问题成为一个可回答的问题，而不仅仅是一个民意调查，我会这样问：
是否有任何第三方报告将 ATI 的 Stream 框架与 NVidia 的 CUDA 框架进行比较（即，不是来自 ATI 或 NVidia 谈论的）自己起来）？
In an effort to make this an answerable question, and not just an opinion poll, I'll ask it like this:
Are there any third-party reports that compare ATI's Stream framework to NVidia's CUDA framework (ie, not from ATI or NVidia talking themselves up)?
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（5）
Tom's Hardware 有一篇文章，ATI Stream：最终，CUDA 有了竞争 (2009-06-15)。
Tom's Hardware has an article, ATI Stream: Finally, CUDA Has Competition (2009-06-15).
我不知道这样的比较，但我会添加一些个人经验。 也许我们可以在这里的用户之间进行适当的比较/优点/缺点/经验/教训。
NVIDIA：
I do not know of such a comparison, but I'll add some personal experience to the mix. Maybe between the users here we can develop a decent comparison/pros/cons/experience/lessons.
NVIDIA:
这种直接比较需要有人构建一个独立于 API 并在 CUDA 和 ATi Stream 上运行的计算测试套件。 人们已经尝试过，但尚未推出独立于平台的解决方案。 可能 GPU 基准测试巨头之一可以完成此任务。 还有其他项目，例如 MAGMA 或 GPUlib，在 Nvidia 的指导下蓬勃发展。 GPGPU 的世界确实非常有党派之争。
Such a direct comparison would require someone to build a computational test suite which is API independent and runs on both CUDA and ATi Stream. People have tried but no platform independent solution is out yet. May be one of the GPU benchmarking powerhouses can get this done. There are other projects such as MAGMA or GPUlib which are blossoming under Nvidia tutelage. The world of GPGPU is very partisan indeed.
从软件的角度来看，这不是苹果与苹果的比较，因为两者的编程模型非常不同，
应用程序可以有效地映射到 Brook+ 而不是 CUDA，反之亦然。 通过编写好的代码来优化硬件的使用会带来很大的不同
From a software perspective, it is not an apples-apples comparison since the programming models of the two are very different,
An application may map efficiently to Brook+ and not to CUDA or vice-versa. Optimum usage of hardware by writing good code makes a big difference
LAMMPS 是最成功的基于 GPU 的分子动力学模拟器之一。 主要开发人员 Axel Kohlmeyer 在本页帖子附件中报告了他们的实验结果。 这些结果是公平且有用的。
LAMMPS is one of the most successful GPU-based molecular dynamic simulators. Axel Kohlmeyer, the main developer, has reported their experimental results on this page in the attachment of the post. These results are fair and useful.