Because of the project needs, so in their own notebook configuration on the Windows GPU version of the Caffe;
Hardware: win10 ; gtx1070 (Computational ability 6.1);
Installation software: cudnn-8.0-windows10-x64-v5.1 ; cuda_8.0.61_win10 ; nugetpackages.zip ; CAF Fe-master;
Can be downloaded on their own website (I also provide Baidu cloud: Link: https://pan.baidu.com/s/1miDu1qo password: w7ja)
Reference link: https://www.cnblogs.com/king-lps/p/6553378.html(mainly in accordance with this and the following two articles to configure)
A Word version of the configuration tutorial : Link: https://pan.baidu.com/s/1hsEiDEG Password: d29u
Http://www.cnblogs.com/zf-blog/p/6139044.html
http://blog.csdn.net/tina_ttl/article/details/51722983
Https://www.2cto.com/kf/201612/575865.html
Detailed steps I will not write first, has been used in Ubuntu Caffe, this configuration is very easy, I only say a few points to note the place:
(1) When installing CUDA, note to uncheck the installation driver compoent option;
(2) Cuda installation is automatically added to the system variables, and after the installation of Cuda it is best to verify (NVCC- V and run DeviceQuery.exe, both better run, the latter I run a flashback, try to resolve the failure, but also did not affect the final installation);
(3) Download after the CUDNN, to extract the three folders to be merged into cuda corresponding folder on the OK;
(4) Download Caffe when the main to download the Windows version of Caffe, is the folder to have Windows this folder;
(5) Compile time does not need to compile all, compile order is Libcaffe, Caffe , if need to compile Python or MATLAB interface;
Then OK, try the simplest mnist on the line;
Caffe GPU version configuration under Windows