【小伟哥AI之路】Docker之参数的调用优化

一、nvidia-docker2 run vs docker run --runtime=nvidia

示例：docker run -d -it -p 8888:8888 tensorflow/tensorflow:2.0.0-gpu-py3-jupyter bash

这样启动容器后是不能调用nvidia显卡的。会提示

在使用nvidia显卡支持需要用如下命令

nvidia-docker run

and

docker run --runtime=nvidia

docker run --runtime nvidia nvidia/cuda:9.0-base nvidia-smi

还可以使用一下命令：

#### Test nvidia-smi with the latest official CUDA image
$ docker run --gpus all nvidia/cuda:9.0-base nvidia-smi

# Start a GPU enabled container on two GPUs
$ docker run --gpus 2 nvidia/cuda:9.0-base nvidia-smi

# Starting a GPU enabled container on specific GPUs
$ docker run --gpus '"device=1,2"' nvidia/cuda:9.0-base nvidia-smi
$ docker run --gpus '"device=UUID-ABCDEF,1"' nvidia/cuda:9.0-base nvidia-smi

# Specifying a capability (graphics, compute, ...) for my container
# Note this is rarely if ever used this way
$ docker run --gpus all,capabilities=utility nvidia/cuda:9.0-base nvidia-smi

都可以设置和启动NV

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.

Existing daemon.json

If you have a custom /etc/docker/daemon.json, the nvidia-docker2 package will override it.
In this case, it is recommended to install nvidia-container-runtime instead and register the new runtime manually.

Default runtime

The default runtime used by the Docker® Engine is runc, our runtime can become the default one by configuring the docker daemon with --default-runtime=nvidia. Doing so will remove the need to add the --runtime=nvidia argument to docker run. It is also the only way to have GPU access during docker build.

Environment variables

The behavior of the runtime can be modified through environment variables (such as NVIDIA_VISIBLE_DEVICES).
Those environment variables are consumed by nvidia-container-runtime and are documented here.
Our official CUDA images use default values for these variables.