Uploading Red Hat GPU Operator Offline Image to Bootstrap Repository¶

This guide explains how to upload an offline image to the bootstrap repository using the nvcr.io/nvidia/driver:525.105.17-rhel8.4 offline driver image for Red Hat 8.4 as an example.

Prerequisites¶

The bootstrap node and its components are running properly.
Prepare a node that has internet access and can access the bootstrap node. Docker should also be installed on this node. You can refer to Installing Docker for installation instructions.

Procedure¶

Step 1: Obtain the Offline Image on an Internet-Connected Node¶

Perform the following steps on the internet-connected node:

Pull the nvcr.io/nvidia/driver:525.105.17-rhel8.4 offline driver image:
```
docker pull nvcr.io/nvidia/driver:525.105.17-rhel8.4
```
Once the image is pulled, save it as a compressed archive named nvidia-driver.tar :
```
docker save nvcr.io/nvidia/driver:525.105.17-rhel8.4 > nvidia-driver.tar
```

Copy the compressed image archive nvidia-driver.tar to the bootstrap node:

scp nvidia-driver.tar user@ip:/root

For example:

scp nvidia-driver.tar root@10.6.175.10:/root

Step 2: Push the Image to the Bootstrap Repository¶

Perform the following steps on the bootstrap node:

Log in to the bootstrap node and import the compressed image archive nvidia-driver.tar :
```
docker load -i nvidia-driver.tar
```

View the imported image:

docker images -a | grep nvidia

Expected output:

nvcr.io/nvidia/driver                 e3ed7dee73e9   1 days ago   1.02GB

Retag the image to correspond to the target repository in the remote Registry repository:
```
docker tag <image-name> <registry-url>/<repository-name>:<tag>
```
Replace with the name of the Nvidia image from the previous step, with the address of the Registry service on the bootstrap node, with the name of the repository you want to push the image to, and with the desired tag for the image.

For example:
```
docker tag nvcr.io/nvidia/driver 10.6.10.5/nvcr.io/nvidia/driver:525.105.17-rhel8.4
```

Push the image to the bootstrap repository:

docker push {ip}/nvcr.io/nvidia/driver:525.105.17-rhel8.4

What's Next¶

Refer to Building Red Hat 8.4 Offline Yum Source and Offline Installation of GPU Operator to deploy the GPU Operator to your cluster.