Building instructions for using CUDA backend

For Ubuntu 17.10 and 18.04 see Troubleshooting section

Should now work on 16.04 and 17.10, tested on fresh 16.04 docker and 17.10 docker

In docker image remove sudo from start of all commands that have it.

TensorFlow build instructions are different and not yet covered here: please add them if you have built TensorFlow support

Docker image: apt update

If you don’t already have, install git: sudo apt install git

Docker 16.04 image needs also gnupg: apt install git gnupg

Docker 17.10 image needs also dirmng: apt install git gnupg dirmngr

Get lc0 source if you don’t already have it. Follow the README.md instructions on how to create a git clone of the repository.

Create /etc/apt/sources.list.d/cuda.list with line:

deb http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64 /

Create /etc/apt/sources.list.d/nvidia-ml.list with line:

deb http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1604/x86_64 /

Add key:

sudo apt-key adv --fetch-keys http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub

Update package list: sudo apt update

Install CUDA 9.2 (will also install newest NVIDIA driver if you don’t already have it: this will likely require reboot before running lc0): sudo apt install cuda

Install cuDNN 7.1: sudo apt-get install libcudnn7 libcudnn7-dev

Install ninja, clang and pip3 (docker 16.04 needs pip3 installation, actual 16.04 includes it):

sudo apt install ninja-build clang python3-pip (UPD: G++ of version 6 or later should also be fine)

Only installing ninja might work: sudo apt install ninja-build

Install/upgrade meson sudo pip3 install --upgrade meson

Build LC0: ./build.sh when you are in lc0/ directory

Resulting lc0 is at lc0/build/release/lc0

Troubleshooting

• If you have an older cuda/cudnn build already you may need to uninstall it.
• I (killerducky) wasn’t able to get this command to work: sudo apt-get install cuda nvidia-cuda-toolkit. I had to follow the instruction here https://docs.nvidia.com/cuda/cuda-installation-guide-linux. After downloading the local .deb file, I ran these commands:

sudo dpkg -i cuda-repo-ubuntu1604-9-1-local_9.1.85-1_amd64.deb 
sudo apt-get update
sudo apt-get install cuda-9.1

• Nvidia cuda does not support gcc later than 6:
  • #error -- unsupported GNU version! gcc versions later than 6 are not supported!.
  • I just ignored them and did this: sudo vim /usr/local/cuda-9.1/bin/../targets/x86_64-linux/include/crt/host_config.h and change it to #if __GNUC__ > 10
  • This is probably only a problem because I’m using Ubuntu 17.10?
  • If there are further issues with Ubuntu 18.04, please edit wiki

Alternative cuDNN install:

Download

• cuDNN v7.1.3 Runtime Library for Ubuntu 16.04 (Deb)
• cuDNN v7.1.3 Developer Library for Ubuntu 16.04 (Deb)
• cuDNN v7.1.3 Code Samples and User Guide for Ubuntu 16.04 (Deb)

from: https://developer.nvidia.com/rdp/cudnn-download

then run

sudo dpkg -i libcudnn7_7.1.3.16-1+cuda9.1_amd64.deb libcudnn7-dev_7.1.3.16-1+cuda9.1_amd64.deb libcudnn7-doc_7.1.3.16-1+cuda9.1_amd64.deb