Theano import error when used with GPU

Question

I have Anaconda Python 2.7.12 installed with numpy 1.11.2 and scipy 0.18.1 versions. I have installed Theano 0.8.2 through conda install. I have added these lines in the .theanorc.txt file:

[global]
floatX = float32 
device = gpu0

[nvcc]
compiler_bindir=C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin`

I have the CUDA Path set in the environment variables. I have also added CUDNN v4.0 to the cuda installation.

However when I do a simple

import theano

in the python interpreter it gives me the following error and falls back to cpu.

`DEBUG: nvcc STDOUT nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
mod.cu
Creating library C:/Users/<USER_NAME>/AppData/Local/Theano/compiledir_Windows-10-10.0.14393-Intel64_Family_6_Model_94_Stepping_3_GenuineIntel-2.7.12-64/cuda_ndarray/cuda_ndarray.lib and object C:/Users/<USER_NAME>/AppData/Local/Theano/compiledir_Windows-10-10.0.14393-Intel64_Family_6_Model_94_Stepping_3_GenuineIntel-2.7.12-64/cuda_ndarray/cuda_ndarray.exp

ERROR (theano.sandbox.cuda): Failed to compile cuda_ndarray.cu: DLL load failed: The specified module could not be found.
WARNING (theano.sandbox.cuda): CUDA is installed, but device gpu0 is not available (error: cuda unavailable)`

The error does not specify which DLL is missing either. I am not sure if this is a bug or a theano installation/config issue or a CUDA installation/config issue.

Thanks

Answer 1

重新安装CUDAv8.0后可以正常工作，很奇怪，因为特定于GPU的CUDA示例（例如devicequery）可以正常工作。