前情提要：
docker是dotCloud 公司为了数据库研发的，所以仅支持CPU和数据交换的部分，nvidia公司觉得这是个好东西，自己派人研发了GPU相关的部分，取名为NVIDIA Container Toolkit，我们可以理解为docker+GPU插件，以后深度学习的环境可以用这个，就不用改中间件版本了。
正文
这个服务器上已经有docker了，但是因为没有NVIDIA Container Toolkit，所以运行GPU的images报错，装上之后正常运行了
报错如下：
docker run --gpus all -it --rm nvcr.io/nvidia/pytorch:24.04-py3
docker: Error response from daemon: could not select device driver "" with capabilities: [[gpu]].
安装代码：
distribution="ubuntu20.04" 这里改成你自己的版本
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
装完了验证的话，可以输入：
dpkg -l | grep nvidia-container-toolkit
显示：
ii nvidia-container-toolkit 1.13.5-1 amd64 NVIDIA Container toolkit
ii nvidia-container-toolkit-base 1.13.5-1 amd64 NVIDIA Container Toolkit Base
然后配置/etc/docker/daemon.json文件，让docker启动时自动调用nvidia驱动
{
"runtimes": {
"nvidia": {
"path": "nvidia-container-runtime",
"runtimeArgs": []
}
},
"default-runtime": "nvidia"
}
重启docker
sudo systemctl restart docker
验证docker是否成功调用GPU
yhp1szh@SZH-C-006RW:/$ docker run --gpus all -it --rm nvcr.io/nvidia/pytorch:24.04-py3
=============
== PyTorch ==
=============
NVIDIA Release 24.04 (build 88113656)
PyTorch Version 2.3.0a0+6ddf5cf
Container image Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2014-2024 Facebook Inc.
Copyright (c) 2011-2014 Idiap Research Institute (Ronan Collobert)
Copyright (c) 2012-2014 Deepmind Technologies (Koray Kavukcuoglu)
Copyright (c) 2011-2012 NEC Laboratories America (Koray Kavukcuoglu)
Copyright (c) 2011-2013 NYU (Clement Farabet)
Copyright (c) 2006-2010 NEC Laboratories America (Ronan Collobert, Leon Bottou, Iain Melvin, Jason Weston)
Copyright (c) 2006 Idiap Research Institute (Samy Bengio)
Copyright (c) 2001-2004 Idiap Research Institute (Ronan Collobert, Samy Bengio, Johnny Mariethoz)
Copyright (c) 2015 Google Inc.
Copyright (c) 2015 Yangqing Jia
Copyright (c) 2013-2016 The Caffe contributors
All rights reserved.
Various files include modifications (c) NVIDIA CORPORATION & AFFILIATES. All rights reserved.
This container image and its contents are governed by the NVIDIA Deep Learning Container License.
By pulling and using the container, you accept the terms and conditions of this license:
https://developer.nvidia.com/ngc/nvidia-deep-learning-container-license
NOTE: The SHMEM allocation limit is set to the default of 64MB. This may be
insufficient for PyTorch. NVIDIA recommends the use of the following flags:
docker run --gpus all --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 ...
root@e09a413296bb:/workspace# nvidia-smi
Thu May 9 05:29:47 2024
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.14 Driver Version: 550.54.14 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 3090 Off | 00000000:17:00.0 Off | N/A |
| 75% 62C P2 203W / 350W | 8385MiB / 24576MiB | 100% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 1 NVIDIA GeForce RTX 3090 Off | 00000000:98:00.0 Off | N/A |
| 43% 41C P8 18W / 350W | 10MiB / 24576MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
+-----------------------------------------------------------------------------------------+
root@e09a413296bb:/workspace# nvcc -C
nvcc fatal : Unknown option '-C'
root@e09a413296bb:/workspace# nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Thu_Mar_28_02:18:24_PDT_2024
Cuda compilation tools, release 12.4, V12.4.131
Build cuda_12.4.r12.4/compiler.34097967_0
root@e09a413296bb:/workspace# python
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> import torchvision
>>> torch.__version__
'2.3.0a0+6ddf5cf85e.nv24.04'
>>> torch.cuda.get_device_name(0)
'NVIDIA GeForce RTX 3090'
>>> torch.backends.cudnn.version()
90100
>>> torchvision.__version__
'0.18.0a0'
然后把代码弄下来测试
因为代理问题，直接从github下载代码太难了，选择手动上传
先把代码传到服务器这个文件夹里
/mnt/workspace/xiebell/pytorch2404/
容器每次关闭，如果用 --rm，是自动删除的，这里我们不要删除
我的代码存进去后启动容器：
docker start 286f0ced9277
docker attach 286f0ced9277
复制服务器文件到容器里：
另开一个terminal
docker cp /mnt/workspace/xiebell/pytorch2404/cuda-samples-master.zip 286f0ced9277:/workspace/xiebell/cuda.zip
显示
Successfully copied 146.6MB to 286f0ced9277:/workspace/xiebell/cuda.zip
此时打开容器查看就有了
解压
至此，文件在容器里已经准备好了，
root@286f0ced9277:/workspace/xiebell/cuda-samples-master# ls
CHANGELOG.md LICENSE README.md Samples_VS2017.sln Samples_VS2022.sln
Common Makefile Samples Samples_VS2019.sln bin
下一步准备编译
编译：
$ cd <sample_dir>
$ make
执行文件
./HSOpticalFlow
结果
root@286f0ced9277:/workspace/xiebell/cuda-samples-master/Samples/5_Domain_Specific/HSOpticalFlow# ./HSOpticalFlow
HSOpticalFlow Starting...
GPU Device 0: "Ampere" with compute capability 8.6
Loading "frame10.ppm" ...
Loading "frame11.ppm" ...
Computing optical flow on CPU...
Computing optical flow on GPU...
L1 error : 0.044308
只能看懂计算误差，看不懂别的
