本文最后更新于：2023年12月5日 下午
Nvidia docker 是nvidia显卡在docker基础上进行封装得到的docker工具，需要电脑中安装Nvidia显卡驱动与docker，配置好Nvidia docker后docker可以使用GPU。本文记录Nvidia docker的安装与使用方法。
环境
- Linux 16.04 64位操作系统
- 显卡驱动 450.80.02
- CUDA 版本 11.0
- docker 安装版本 19.03.4
安装过程
1
2
3
4
5
6
| curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | \
sudo apt-key add -
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | \
sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt-get update
|
- 安装nvidia-docker2 载入docker 配置
1
2
| sudo apt-get install -y nvidia-docker2
sudo pkill -SIGHUP dockerd
|
1
| docker run --runtime=nvidia --rm nvidia/cuda:11.0.3-base-${DIST} nvidia-smi
|
其中 DIST 为 ubuntu20.04, ubuntu18.04, centos7 其中之一。
nvidia/cuda:11.0-base 镜像已经被移除
此时会显示出显卡信息，说明nvidia docker成功创建并在内部正确执行了 nvidia-smi 命令。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
| $ sudo docker run --runtime=nvidia --rm nvidia/cuda:11.0.3-base-ubuntu20.04 nvidia-smi
Unable to find image 'nvidia/cuda:11.0.3-base-ubuntu20.04' locally
11.0.3-base-ubuntu20.04: Pulling from nvidia/cuda
96d54c3075c9: Pull complete
59f6381879f6: Pull complete
655ed0df26cf: Pull complete
848b95ad96b5: Pull complete
e43c2058e496: Pull complete
Digest: sha256:c8269d6967e10940c368ea24fb8086cb21471cb8fefc66861d72f74f0c67e904
Status: Downloaded newer image for nvidia/cuda:11.0.3-base-ubuntu20.04
Tue Dec 5 08:33:54 2023
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.129.03 Driver Version: 535.129.03 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 NVIDIA GeForce RTX 3080 Off | 00000000:01:00.0 On | N/A |
| 30% 27C P8 20W / 320W | 503MiB / 10240MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
+---------------------------------------------------------------------------------------+
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
| $ sudo apt show nvidia-docker2
->
Package: nvidia-docker2
Version: 2.5.0-1
Priority: optional
Section: utils
Maintainer: NVIDIA CORPORATION <cudatools@nvidia.com>
Installed-Size: 27.6 kB
Depends: nvidia-container-runtime (>= 3.4.0), docker-ce (>= 18.06.0~ce~3-0~ubuntu) | docker-ee (>= 18.06.0~ce~3-0~ubuntu) | docker.io (>= 18.06.0)
Breaks: nvidia-docker (<< 2.0.0)
Replaces: nvidia-docker (<< 2.0.0)
Homepage: https://github.com/NVIDIA/nvidia-docker/wiki
Download-Size: 5,840 B
APT-Manual-Installed: yes
APT-Sources: https://nvidia.github.io/nvidia-docker/ubuntu16.04/amd64 Packages
Description: nvidia-docker CLI wrapper
Replaces nvidia-docker with a new implementation based on
nvidia-container-runtime
N: There are 50 additional records. Please use the '-a' switch to see them.
|
出现类似信息说明安装成功
使用nvidia docker镜像
- 之前不是同一个驱动版本下创建的镜像在新驱动下的nvidia docker中可能找不到nvidia-smi命令
- 为了在新的docker下使用gpu，我迂回地使用测试镜像作为初始镜像
1
| nvidia-docker run -it --name first_container nvidia/cuda:11.0.3-base-${DIST} /bin/bash
|
其中 DIST 为 ubuntu20.04, ubuntu18.04, centos7 其中之一。
这样便将测试镜像创建出了可以随时访问修改的容器，在该容器基础上保存镜像即可。
1
| docker commit -m "nvidia docker image init" first_container my_image:1.0
|
参考资料
文章链接：
https://www.zywvvd.com/notes/tools/docker/nvidia-docker-install/
