Installing NVIDIA Container Toolkit and Docker

You can install NVIDIA Container Toolkit on Ubuntu and Red Hat® Enterprise Linux® operating systems to enable support for GPU devices in the Docker containers. You can install Docker on Ubuntu or Red Hat Enterprise Linux 7 or operating systems.

About this task

NVIDIA Container Toolkit

To install the NVIDIA Container Toolkit on Ubuntu and Red Hat Enterprise Linux operating systems, see the Installing the NVIDIA Container Toolkit documentation on the NVIDIA website.

Note:

Starting in Maximo® Visual Inspection Edge 8.8, IBM POWER8 is no longer supported.

Starting in Maximo Visual Inspection Edge 8.9, IBM Power® Systems are no longer supported.

Docker

To install Docker on Red Hat Enterprise Linux, use the following instructions from the Install Docker Engine documentation on the Docker website:

What to do next

NVIDIA Container Toolkit

When the NVIDIA Container Toolkit is installed on the Red Hat Enterprise Linux 8 operating system, complete the following:

Generate a Container Device Interface specification.

For more information, see Generating a CDI specification on the NVIDIA website.

Docker

When Docker is installed on Ubuntu or Red Hat Enterprise Linux 7 operating system, complete the following steps:

Open the Docker daemon configuration file by entering the following command:

/etc/docker/daemon.json

Configure the default runtime to ensure that locally deployed models run correctly on GPUs.

The configuration file contains the following NVIDIA runtime:


{   
 "runtimes" : { 
      "nvidia" : { 
         "path" : "/usr/bin/nvidia-container-runtime", 
         "runtimeArgs" : [] 
      } 
   }
 }

To make NVIDIA runtime the default runtime, add the default-runtime option to the configuration file.
Enable log file rotation by updating the log-driver and log-opts options in the configuration file.

Maximo Visual Inspection Edge  containers write log messages to the Docker container logs, which can become large in production systems.

For example, after you make these updates, the daemon configuration file might contain the  following configuration:
```
{    "runtimes" : { 
      "nvidia" : { 
         "path" : "/usr/bin/nvidia-container-runtime",
          "runtimeArgs" : [] 
      } 
   }, 
   "default-runtime" : "nvidia", 
   "log-driver": "json-file", 
   "log-opts": { 
      "max-size": "10m", 
       "max-file": "20" 
   }
 }
```
Save your changes and restart the Docker daemon by  running the following command:
```
sudo systemctl restart docker
```