Notes on tensorflow-GPU Installation

Install the SDK in the correct order and strictly install the specified version.

1. download and install the strict version of Cuda and cudnn. Other versions do not work. For example, if 9.0 is required, you cannot set 9.1. Https://www.tensorflow.org/install/install_windows

1.1. Delete c: \ Program Files \ NVIDIA Corporation \ installer2 before installing 9.0 pattern. Otherwise, the system will crash.

1.2. After cudnn is installed, check whether c: \ Program Files \ nvidia gpu computing toolkit \ Cuda \ v9.0 \ bin and libnvvp exist in % PATH %.

And % cuda_path % = c: \ Program Files \ nvidia gpu computing toolkit \ Cuda \ v9.0

And % cuda_path_v9_0 % = c: \ Program Files \ nvidia gpu computing toolkit \ Cuda \ v9.0

2. Install the video card driver. This can be installed with the latest one.

3. Pip install tensorflow-GPU. Do not install a non-"-GPU" CPU version. If you have installed the GPU, uninstall the GPU.

3. Pip uninstall protobuf, pip install protobuf = 3.6.0 (same as your Python version, 3.6.1 does not work)

4. Perform Python verification. Import tensorflow as TF; sess = TF. Session (Config = TF. configproto (log_device_placement = true ));

Notes on tensorflow-GPU Installation