has been in Linux under the Theano,gpu with a good match. Need to work under Windows last week, so toss for a week, just inexplicably to match the GPU.
First on a Theano successful use of the GPU screenshot
Here is my experience in configuring Theano:
It's basically two steps away:
1. Installation Theano
2. Installation Cuda
Note that under Win7 64, Python and Cuda should be unified, either with 32-bit or 64-bit.
One of the most troublesome possibilities under Windows is the installation of Theano, which is http://deeplearning.net/software/theano/install.html#windows step-by-step according to the official documentation steps.
Install Theano:
First of all, to install Theano dependencies, the most important is the numpy,scipy,nose. At first I installed the Numpy,scipy,theano separately, it is easy to install the Theano, but after the Cuda, dead and dead can not use the GPU; later try to install Theano dependencies with Pythonxy, then download the Theano source code, with Python setup.py install installed, but the GPU is still not used, a whole week ah ...
The first way I install the official website is enthought Python distribution (EPD). Install Theano dependencies, then install the Theano with Pip, and finally the GPU is available.
Install Cuda:
Cuda is using Cuda_5.0.35_winvista_win7_win8_general_32-3.msi, download the address
Https://developer.nvidia.com/cuda-toolkit-archive
vs I use 2008, recommend not to use vs2012, or more trouble
Here is my Theano profile, like the official web, you can use the
. Theanorc.txt:
My Baidu cloud inside has the related software package, like CUDA5.0,EPD and so on