CUDA 8.0 + cuDNN v6.0 installation (Linux)
Low version of tensorflow, for example, 1.4.0 calling GPU requires low version of CUDA 8.0 and cuDNN v6.0
The corresponding relationship between tensorflow and CUDA of each version is as follows:
download
Download CUDA8.0 corresponding to Linux and cuDNN6.0 corresponding to CUDA8.0: CUDA 8.0 NVIDIA official website ; cuDNN 6.0 for CUDA 8.0 (registration required) NVIDIA official website ;
installation
Refer to the official website Guide CUDA 8.0 for Linux ; cuDNN 6.0 for CUDA 8.0 for Linux ;
other reference blogs CUDA8.0 and 9.0 version switch ; cuDNN 6.0 installation ; the
author completed the installation according to other reference blogs.
Verify tensorflow-gpu availabel
When the install package Pycharm direct attention to install tensorfolw-gpu
reference verification code tensorflow this case is whether to use the GPU ;