Hii Doug
When I tried to run this script my screen was frozen for about15 mins. Does
this mean that there is something wrong in my CuDNN installation? Or should
I try running the script again?
On Thursday, April 7, 2016 at 11:06:53 AM UTC-5, Doug wrote:
>
> Theano can have a hard time detecting CuDnn on windows if you rely on
> config flags in your theanorc file to get everything working. If you are
> confident you have installed CuDnn correctly on your system, you can force
> your script to try and use it by manually setting
> theano.sandbox.cuda.dnn.dnn_available.avail to True. See the below script
> to quickly test if CuDnn is working properly.
>
> import theano
> import numpy as np
>
>
> if __name__ == '__main__':
> theano.sandbox.cuda.dnn.dnn_available.avail=True
>
> img = theano.tensor.tensor4()
>
> out = theano.sandbox.cuda.dnn.dnn_pool(img, (2,2))
>
> f = theano.function([img],out)
> d = np.random.rand(10,10,10,10).astype('float32')
>
> print f(d).shape
>
>
>
>
> On Thursday, April 7, 2016 at 9:00:06 AM UTC-4, Kristofer Krus wrote:
>>
>> I have installed Theano, CUDA and CuDNN, and Theano can find CUDA, but it
>> can't find CuDNN. I have added exported CUDNN_PATH as 'C:\Program1\cudnn',
>> which is where the folder in which I have installed CuDNN. I have also
>> followed the instructions on this page
>> <http://deeplearning.net/software/theano/library/sandbox/cuda/dnn.html>
>> for installing cuDNN by running
>>
>> export LD_LIBRARY_PATH=path_to_CUDNN_folder/lib/x64:$LD_LIBRARY_PATH
>> export CPATH=path_to_CUDNN_folder/include:$CPATH
>> export LIBRARY_PATH=path_to_CUDNN_folder/lib/x64:$LD_LIBRARY_PATH
>>
>>
>> in git bash. The instructions on the page I linked to are already
>> modified as they stand here, since the CuDNN folder doesn't contain a lib64
>> folder in my installation on Windows 7, but a lib/x64 folder (also, I don't
>> have cuDNN installed under my /home/user folder). However, I don't know
>> whether these variables are used in Windows, as modifying them like
>> described on the page doesn't work. When I run the python script defined
>> here
>> <http://deeplearning.net/software/theano/install_windows.html#configuring-theano>,
>>
>> I get the output
>>
>> Using gpu device 0: GeForce GTX 750 (CNMeM is enabled with initial size:
>> 95.0% of memory, CuDNN not available)
>> NP time: 0.077008[s], theano time: 0.072007[s] (times should be close
>> when run on CPU!)
>> Result difference: 0.019287
>>
>> which indicates that Theano doesn't find CuDNN. It also says that it's
>> using gpu device 0, but 0.072007 seconds with Theano doesn't seem much
>> faster than 0.077008 seconds with Numpy, so does Theano really use the GPU?
>>
>> I have also copied all files in the CuDNN folder to their corresponding
>> positions in the CUDA folder, but that hasn't helped either.
>>
>> Anyone knows why Theano can't find CuDNN or what I have to do to make
>> Theano find CuDNN?
>>
>> Regards
>>
>> Kristofer
>>
>
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
