结论

凡是使用了命令nvidia docker或者在docker中引入了--runtime=nvidia参数的都是过时教程，最新方法只需要下载nvidia-container-toolkits，在docker中引入--gpus参数即可

1. nvidia docker

nvidia docker是NVIDIA第一代支持docker容器内使用GPU资源的项目。运行时用nvidia-docker命令。

根据nvidia docker在github上的描述，已经不再使用了。

The nvidia-docker wrapper script that was included in this repository is no longer included in the package and a configuration specific to the target container engine (e.g. Docker, Containerd, Cri-o, or Podman) is suggested instead.

2. nvidia docker2

nvidia docker2是NVIDIA第二代支持docker容器内使用GPU资源的项目。运行时用nvidia-docker命令，且需要指定参数--runtime=nvidia.

根据github wiki的描述，一代和二代之间有如下兼容性。

Backward compatibility To help transitioning code from 1.0 to 2.0, a bash script is provided in /usr/bin/nvidia-docker for backward compatibility. It will automatically inject the --runtime=nvidia argument and convert NV_GPU to NVIDIA_VISIBLE_DEVICES.

也就是说，在二代中，既可以使用nvidia docker命令，这会自动引入参数--runtime=nvidia也可以使用docker命令，手动指定参数--runtime=nvidia 如帖子总结：

docker run --runtime=nvidia is only available since nvidia-docker v2. Both commands are equivalent with nvidia-docker v2, the former is a script provided for backward compatibility with nvidia-docker v1.

ps: 笔者上一次使用docker gpu 还停留在 nvidia docker2 的阶段. 技术果然在飞速进步啊.

3. nvidia-container-toolkits

根据github这是最新的支持方案，如帖子描述，nvidia docker2 被Nvidia container toolkits取代，无需指定--runtime参数，只需要传递--gpus参数

root@docker2-1:/# docker run --rm --gpus all ubuntu nvidia-smi
Tue Sep 10 08:31:49 2024       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.107.02             Driver Version: 550.107.02     CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| 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 2080 Ti     Off |   00000000:03:00.0 Off |                  N/A |
|  0%   41C    P0             57W /  300W |       1MiB /  11264MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   1  NVIDIA GeForce RTX 3060 Ti     Off |   00000000:04:00.0 Off |                  N/A |
|  0%   49C    P0             46W /  220W |       1MiB /   8192MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

原文链接：https://blog.csdn.net/yjy420/article/details/132305356