eric response

output.txt

I can test your workflow if you give me the steps. These are the steps I conducted to verify the environment.
 
salloc –N 1 –p gpu
# Gives me node084
ssh node084
module load cuda75/toolkit/7.5.18
module load python/anaconda
# Testing out CUDA (optional and not required, but this is how you would do it)q
cd $CUDA_SDK
./verify_cuda75.sh
# Some failed, but most succeed so I know the GPU is running and accessible
 
 
# Testing out theano
Used the http://deeplearning.net/software/theano/tutorial/using_gpu.html#testing-theano-with-gpu script to test whether we’re using the CPU or GPU.  Despite $CUDA_ROOT being specified as documentation instructs the above script still used the CPU.
Created the .theanorc in my home directory as:
[global]
device = gpu
floatX = float32
After this file’s creation running the test script shows GPU is being used.
[eric@node084 theano]$ python test.py
Using gpu device 0: Tesla K20m (CNMeM is disabled, CuDNN not available)
[GpuElemwise{exp,no_inplace}(<CudaNdarrayType(float32, vector)>), HostFromGpu(GpuElemwise{exp,no_inplace}.0)]
Looping 1000 times took 0.754345 seconds
Result is [ 1.23178029  1.61879349  1.52278066 ...,  2.20771813  2.29967761
  1.62323296]
Used the gpu
 
Looking at your .theanorc you already have those lines. Could be the root line you have. The root that is defined there should be /path/to/cuda/root, which when loading the module is /cm/local/apps/cuda/libs/current not the cuda files in your conda install.
 
Eric Borenstein