TensorFlow 1.8, Ubuntu 16.04, Cuda 9.0, nvidia-390, installation Pit Guide.

By TensorFlow 1.8, Ubuntu 16.04, Cuda 9.0, nvidia-390 tortured for 5 days, finally on the pit, leaving a guide for the benefit of posterity.

1. Find out the dependencies first:

TensorFlow 1.8 relies on Cuda 9.0,cuda 9.0 dependent nvidia-390.

2. Pit:

Only nvidia-384,nvidia-390 newness is not supported in Ubuntu 16.04 source.

What to do?

If you're not in the kingdom, you can install nvidia-390 in your personal source

• sudo add-apt-repository Ppa:graphics-drivers/ppa
• sudo apt update
• sudo apt install nvidia-390

If you are very slow in the celestial PPA, you can only download it manually from here:

http://ppa.launchpad.net/graphics-drivers/ppa/ubuntu/pool/main/n/nvidia-graphics-drivers-390/nvidia-390_ 390.48-0ubuntu0~gpu16.04.3_amd64.deb

3. Installation order, it is critical to install the first Cuda and then nvidia-390 order, otherwise it will cause loop login Loop login cannot enter the desktop:

• 1. Refer to NVIDIA documentation installation Cuda9.0:https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#ubuntu-installation
• 2. Install the downloaded nvidia-390 deb file: Dpkg-i path/to/nvidia-390_390.48-0ubuntu0~gpu16.04.3_amd64.deb
• 3. Install the dependency package: sudo apt-get install-f. The reason is if the dpkg install a package and package dependency are not satisfied, it leaves the unconfigured Ckage is considered as broken. sudo apt-get install -fcommand tries to fix this broken package by installing the missing dependency.

4. Finish.

TensorFlow 1.8, Ubuntu 16.04, Cuda 9.0, nvidia-390, installation Pit Guide.