处理器的千兆浮点运算
我发现我的计算机具有 NVIDIA CUDA 技术,并且我想测量 CPU 和 GPU 的处理能力。
我不想寻找一个程序来执行此操作,而是想更深入地了解它是如何工作的。我需要什么样的代码(C/C++)?
I discovered my computer has NVIDIA CUDA Technology and I want measure the power of processing, in CPU and GPU.
Instead of searching for a program to do this, I want have a deeper understanding of how it works. What kind of code (C/C++) I need?
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(1)
您需要下载CUDA SDK,可能还有一些示例并开始您的GPU编程。然后你应该阅读一些关于什么是 FLOPS 以及如何正确测量它们的信息(从这里开始 什么是 FLOP/s,它是性能的一个很好的衡量标准吗? 然后进行一些谷歌搜索)。然后进行测试:)
PS:我想警告一下 CUDA 及其性能。有关 CUDA 和 GPU 的营销材料承诺速度至少比任何 CPU 高 10 倍,您的测试可能会得到这些值。但它们并不是纯粹 FLOPS,GPU 并不是像 CPU 那样的通用计算设备,并且您不会在每种算法上获得相同的结果。
You need to download CUDA SDK, some samples maybe and start your GPU programming. Then you should read some info about what are FLOPS actually and how to measure them correctly (start here What is FLOP/s and is it a good measure of performance? then some googling). Then do your tests :)
PS: I want to warn about CUDA and its performance. Marketing materials about CUDA and GPUs promise minimum x10 higher speeds than of any CPUs and your tests might get these values. But they are not pure FLOPS, GPU is not that universal computational device as CPU is and you will not achieve same results on every algorithm.