Installing TensorFlow: Difference between revisions

Revision as of 10:36, 16 November 2017

First install CUDA
- Apparently CUDA 9.0 won't fly. Be sure to use CUDA 8.0 until 9.0 is validated. Sigh.
- CUDA 8.0 has been saved to the UBFS folder. Install it using sudo apt-get install cuda-8-0. The logic behind installing this way is apparent when you look at Table 4 in the current iteration of the installation guide for CUDA 9.0

Modify .bashrc to include CUDA in your PATH and LD_LIBRARY_PATH
- export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64:$LD_LIBRARY_PATH
- export PATH=/usr/local/cuda-9.0/bin:$PATH
- export CUDA_HOME=/usr/local/cuda-9.0
Install cuDNN
Now install TensorFlow

@@ Line 1: / Line 1: @@
 #First  [http://docs.nvidia.com/cuda/cuda-installation-guide-linux/#axzz4VZnqTJ2A install CUDA]
 #*[https://devtalk.nvidia.com/default/topic/1026198/cuda-9-0-importerror-libcublas-so-8-0/ Apparently CUDA 9.0 won't fly]. Be sure to use CUDA 8.0 until 9.0 is validated. Sigh.
+#*CUDA 8.0 has been saved to the UBFS folder. Install it using <code>sudo apt-get install cuda-8-0</code>. The logic behind installing this way is apparent when you look at [http://docs.nvidia.com/cuda/cuda-installation-guide-linux/#package-manager-metas Table 4] in the current iteration of the installation guide for CUDA 9.0
 #Modify .bashrc to include CUDA in your PATH and LD_LIBRARY_PATH
 #*<code>export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64:$LD_LIBRARY_PATH</code>