I have been using theano in linux, and the gpu is very good. I had to work in windows last week, so I had a hard time working for a week. I just had a gpu configuration.
First, the gpu is successfully used by theano.
The following is my experience in theano Configuration:
There are basically two steps:
1. Install theano
2. Install cuda
Note that python and cuda must be unified in win7 64-Bit mode, either using 32-bit or 64-bit.
In windows, the most troublesome thing is to install theano. Follow the steps in the official documentation: http://deeplearning.net/software/theano/install.html#windows.
Install theano:
First, install theano dependencies. The most important ones are numpy, scipy, and nose. At the beginning, I installed numpy, scipy, and theano separately. It was easy to install theano, but after installing cuda, I couldn't use the gpu. Later I tried to install theano dependencies with pythonxy, download theano source code and use python setup. py install installation, but the gpu still cannot be used. It took a whole week...
Next I will install the first method on the official website, that is, Enthought Python Distribution (EPD). Install the dependencies of theano, then install theano with pip, and finally the gpu can be used.
Install cuda:
Cuda uses cuda_5.0.35_winvista_win7_win8_general_32-3.msi,
Https://developer.nvidia.com/cuda-toolkit-archive
Vs I use 2008. We recommend that you do not use vs2012. Otherwise, it will be troublesome.
The following is my theano configuration file, which is available on the official website.
.Theanorc.txt:
My Baidu cloud has related software packages, such as cuda5.0 and EPD.