Reputation: 1177
i am running a gtx 980ti on ubuntu 16.04, installed cuda 7.5 via deb file and cupied the cudnn files into /usr/local/cuda-7.5 ... .
I edited my bashrc file with the following:
export CUDA_HOME=/usr/local/cuda-7.5 export LD_LIBRARY_PATH=${CUDA_HOME}/lib64
PATH=${CUDA_HOME}/bin:${PATH} export PATH
Now when i run a model, i get the following :
I tensorflow/stream_executor/dso_loader.cc:102] Couldn't open CUDA library libcublas.so. LD_LIBRARY_PATH:
I tensorflow/stream_executor/cuda/cuda_blas.cc:2289] Unable to load cuBLAS DSO.
I tensorflow/stream_executor/dso_loader.cc:102] Couldn't open CUDA library libcudnn.so. LD_LIBRARY_PATH:
I tensorflow/stream_executor/cuda/cuda_dnn.cc:2259] Unable to load cuDNN DSO
I tensorflow/stream_executor/dso_loader.cc:102] Couldn't open CUDA library libcufft.so. LD_LIBRARY_PATH:
I tensorflow/stream_executor/cuda/cuda_fft.cc:343] Unable to load cuFFT DSO.
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:102] Couldn't open CUDA library libcurand.so. LD_LIBRARY_PATH:
I tensorflow/stream_executor/cuda/cuda_rng.cc:333] Unable to load cuRAND DSO.
Does anyone see what i am doing wrong here and or knows how to fix it. All suggestions are highly appreciated :)
Upvotes: 0
Views: 579
Reputation: 4648
You must also install the package nvidia-cuda-dev
- this will pull in all of the other dependencies such as libcublas, libcufft and libcurand, as well as their plain .so
symbolic links.
On my Ubuntu 16.04 system, these CUDA 7.5 libraries all ended up in /usr/lib/x86_64-linux-gnu
so I did not even need the LD_LIBRARY_PATH
setting.
Furthermore, I installed the two packages libcudnn5_5.1.3-1+cuda7.5_amd64.deb
and libcudnn5-dev_5.1.3-1+cuda7.5_amd64.deb
via my NVIDIA Developer account.
Now when I start ipython in a TensorFlow 0.10 configured environment, with LD_LIBRARY_PRELOAD
empty, I see:
In [1]: import tensorflow as tf
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcublas.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcudnn.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcufft.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcurand.so locally
Joy!
Upvotes: 1