[Ubuntu18.10] 给docker容器加上gpu加速

1)

• Make sure an nvidia driver is installed on the host system

2)

• Follow the steps here to setup the nvidia container toolkit

NVIDIA 容器工具包允许用户构建和运行 GPU 加速的 Docker 容器。该工具包包括一个容器运行时库和实用程序，可自动配置容器以利用 NVIDIA GPU。

安装container toolkit教程:(ubuntu18.10会特有的问题见下面的帖子)

Installation Guide — NVIDIA Cloud Native Technologies documentation

1​​​​​​解决 Ubuntu18.10 无法安装 Nvidia-docker2_吴中骊的博客-CSDN博客_ubuntu无法安装docker

测试nvidia container toolkit安装成功

sudo docker run --rm --gpus all nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04 nvidia-smi

3)做一个含cuda和cudnn的image

• Make sure cuda, cudnn is installed in the image

 4)运行image为容器

• Run a container with the --gpus flag (as explained in the link above)

docker run --gpus all -it -d --name py0331 pytorch:latest /bin/bash 

验证pytorch已经支持gpu加速:

Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
True
>>> torch.cuda.device_count()
1
>>> torch.cuda.current_device()
0
>>> torch.cuda.get_device_name(0)
'GeForce GTX 1080 Ti'
>>> quit()