Rapids / docker: 无法 select 具有功能的设备驱动程序“”：[[gpu]]

Question

我是 Rapids 的新手，很少有 conda 的良好体验。所以我正在尝试使用容器化版本。我是 Docker 的新手，未知数的组合让我无法解决问题。

我有一个 Ubuntu 18.04 服务器，

# uname -v
#30~18.04.1-Ubuntu SMP Fri Jan 17 06:14:09 UTC 2020

我在上面安装了 Docker

的新版本

# apt-get install docker docker-ce docker-ce-cli containerd.io
# docker --version
Docker version 19.03.8, build afacb8b7f0

本机安装了cuda v10.2

# nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Wed_Oct_23_19:24:38_PDT_2019
Cuda compilation tools, release 10.2, V10.2.89

和Python v3.6.9

# python3 --version
Python 3.6.9

如 NVIDIA Container Toolkit Quickstart 部分所示，我将 nvidia-docker 列表安装到 /etc/apt/sources.list.d/

# curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
# curl -s -L https://nvidia.github.io/nvidia-docker/ubuntu18.04/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list

明确用 ubuntu18.04 代替 $distribution，因为那是 Ubuntu equivalent for Linux Mint 19.3.

按照 RAPIDS - Open GPU Data Science 中的启动容器和笔记本服务器说明，我拉取了 0.13-cuda10.2-runtime-ubuntu18.04-py3.6 运行时。

# docker pull rapidsai/rapidsai:0.13-cuda10.2-runtime-ubuntu18.04-py3.6

好久好久，好几GB了，好像都OK了。（没有警告或错误消息。）此外，该图像似乎已在 Docker.

中注册

# docker images -a
REPOSITORY          TAG                                       IMAGE ID            CREATED             SIZE
rapidsai/rapidsai   0.13-cuda10.2-runtime-ubuntu18.04-py3.6   c7440af853b5        4 days ago          9.26GB
rapidsai/rapidsai   cuda10.2-runtime-ubuntu18.04-py3.6        c7440af853b5        4 days ago          9.26GB

但是，我接下来尝试启动笔记本服务器：

# docker run --gpus all --rm -it -p 8888:8888 -p 8787:8787 -p 8786:8786 \
       rapidsai/rapidsai:cuda10.0-runtime-ubuntu18.04-py3.6
docker: Error response from daemon: could not select device driver "" with capabilities: [[gpu]].

这似乎令人惊讶，因为检测到两个 GTX 1080 Ti GPU

# nvidia-smi
Fri May  8 16:41:57 2020       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.33.01    Driver Version: 440.33.01    CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 108...  On   | 00000000:08:00.0 Off |                  N/A |
| 21%   38C    P8    10W / 250W |      1MiB / 11178MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 108...  On   | 00000000:42:00.0 Off |                  N/A |
| 23%   42C    P8    10W / 250W |      1MiB / 11177MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

清理后

# docker system prune -a
# apt-get purge docker docker-engine docker.io containerd runc

我重新安装了docker并再次拉取了rapidsai镜像。结果不变。

是否与NVIDIA驱动版本：440.33.01有冲突？

有什么建议吗？

Answer 1

感谢您试用 RAPIDS。您碰巧安装了 nvidia-container-toolkit 吗？ https://github.com/NVIDIA/nvidia-docker#quickstart. I didn't see that in your steps and missing it could cause that issue. It's in our prerequisites on https://rapids.ai/start.html

Answer 2

来自NVIDIA CUDA/WSL 2 documentation：

Use the Docker installation script to install Docker for your choice of WSL 2 Linux distribution. Note that NVIDIA Container Toolkit does not yet support Docker Desktop WSL 2 backend.

Answer 3

我只是按照中的步骤操作；它工作正常：

要卸载以前的 nvidia-docker 软件包，请发出这些命令：

[user@gpu1 ~]# docker volume ls -q -f driver=nvidia-docker | xargs -r -I{} -n1 docker ps -q -a -f volume={} | xargs -r docker rm –f
[user@gpu1 ~]# sudo apt-get remove nvidia-docker

要安装 NVIDIA-GPU Docker 容器工具包，您首先需要添加包存储库：

user@ubuntu-gpu1:~# distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
user@ubuntu-gpu1:~# curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
user@ubuntu-gpu1:~# curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
user@ubuntu-gpu1:~# sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
user@ubuntu-gpu1:~# sudo systemctl restart docker

然后使用最新的官方 CUDA 映像验证 nvidia-smi 安装：

user@ubuntu-gpu1:~# sudo docker run -it --rm --gpus all nvidia/cuda:9.0-base nvidia-smi

Answer 4

试试这个

sudo apt install -y nvidia-docker2
sudo systemctl daemon-reload
sudo systemctl restart docker

Rapids / docker: 无法 select 具有功能的设备驱动程序“”：[[gpu]]

Rapids / docker: could not select device driver "" with capabilities: [[gpu]]

docker

gpu

nvidia

rapids