Fail to find the dnn implementation.

Accelerated Computing CUDA CUDA Setup and Installation
1

I have recently been trying to setup my new gtx 1080 with keras 2.1.4 tensorflow 1.50, cudnn 7.05 and cuda toolkit 9. Everything appears to be working except for when I try to use some of the cudnn optimized operations in keras like cudnngru and cudnnlstm. I have not tried to the other cudnn operations, but I can get the basic starter code setup where they open a session and then just print hello tensorflow.

I am getting this error
UnknownError (see above for traceback): Fail to find the dnn implementation.
[[Node: cu_dnngru_1/CudnnRNN = CudnnRNN[T=DT_FLOAT, direction=“unidirectional”, dropout=0, input_mode=“linear_input”, is_training=true, rnn_mode=“gru”, seed=87654321, seed2=0, _device=“/job:localhost/replica:0/task:0/device:GPU:0”](cu_dnngru_1/transpose, cu_dnngru_1/ExpandDims_1, cu_dnngru_1/Const_1, cu_dnngru_1/concat)]]
[[Node: loss_1/mul/_321 = _Recvclient_terminated=false, recv_device=“/job:localhost/replica:0/task:0/device:CPU:0”, send_device=“/job:localhost/replica:0/task:0/device:GPU:0”, send_device_incarnation=1, tensor_name=“edge_1145_loss_1/mul”, tensor_type=DT_FLOAT, _device=“/job:localhost/replica:0/task:0/device:CPU:0”]]

and I cant seem to find any other information on this error anywhere online. Any troubleshooting steps would be appreciated. I followed along with a bunch of different tensorflow install videos but I cant seem to get it to work.

2

2018-03-03 16:25:15.383774: E C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\35\tensorflow\stream_executor\cuda\cuda_dnn.cc:378] Loaded runtime CuDNN library: 7101 (compatibility version 7100) but source was compiled with 7003 (compatibility version 7000). If using a binary install, upgrade your CuDNN library to match. If building from sources, make sure the library loaded at runtime matches a compatible version specified during compile configuration.

Also just found this additional information. Will look into this further but not positive what it means

3

Downgraded to 7.0.4 and installed tensorflow 1.6 and it is working now.

4

Thanks for your notes. Worked for me too.

5

Hi, upgrading my tensorflow to the newest version (1.9) solved my problem :)

pip uninstall tensorflow protobuf --yes
find $CONDA_PREFIX -name “tensorflow” | xargs -Ipkg rm -rfv pkg
pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.9.0-cp36-cp36m-linux_x86_64.whl --no-cache-dir

6

In my case it was older CuDNN version: 7 when TF requiered >=7.2

7

I was using pip to install tensorflow and keras
after multiple attempts of failure, I uninstalled these and did a conda install


conda install tensorflow
conda install keras

This worked
:)

8

For me executing following lines just after tensorflow import worked for me:

physical_devices = tf.config.list_physical_devices(‘GPU’)
tf.config.experimental.set_memory_growth(physical_devices[0], True)

Setup:
Tf = 2.2.0
cuda toolkit = 10.1
cudNN = 7.6

1 Like