bangbangbang
bangbangbang

Reputation: 41

tensorflow: undefined symbol: cudnnSetRNNDescriptor_v6

when I finished installing tensorflow (GPU_support,linux 14.04,python3.4) with virtualenv environment,under the instructions of the official website, i validated the installation with the command :python; import tensorflow; but there is an error:

Upvotes: 4

Views: 2433

Answers (1)

9th Dimension
9th Dimension

Reputation: 305

I had this same error, and so hopefully this solution will work for you...

What version of CuDNN are you using? I found that versions of tensorflow-gpu > 1.2 would fail to load while I had CuDNN v5.1.10 installed.

At the time I'm writing this, Tensorflow docs say you must have CuDNN v6. https://www.tensorflow.org/install/install_linux#nvidia_requirements_to_run_tensorflow_with_gpu_support

So I just removed the old CuDNN binaries and headers:

$ sudo rm /usr/local/cuda/include/cudnn.h

$ sudo rm /usr/local/cuda/lib64/libcudnn.so

$ sudo rm /usr/local/cuda/lib64/libcudnn.so.5.1.10

and any other remnants of v5.1.10. Then just download and copy the new v6 headers and binaries into those same locations.

This is a useful command to check your CuDNN version:

$ cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

Upvotes: 3

Related Questions