Installing NVIDIA drivers on a new VM instance can be
time consuming, especially if you are creating numerous images. One way to
prevent this is to create your own image that is based on one of the
Deep Learning VM images, but that already has the NVIDIA drivers
preinstalled.

This topic describes how to create a new image based on an existing
Deep Learning VM image.

## Create a new instance

First, follow the instructions in one of the following topics to create a new
instance. Be sure to include at least one GPU in your new instance.

- [Creating a TensorFlow Deep Learning VM
  Instance](https://docs.cloud.google.com/deep-learning-vm/docs/tensorflow_start_instance)
- [Creating a PyTorch Deep Learning VM
  Instance](https://docs.cloud.google.com/deep-learning-vm/docs/pytorch_start_instance)
- [Creating a Deep Learning VM Instance from the Command
  Line](https://docs.cloud.google.com/deep-learning-vm/docs/cli)
- [Creating a Deep Learning VM Instance from the Cloud
  Marketplace](https://docs.cloud.google.com/deep-learning-vm/docs/cloud-marketplace)

## Verify NVIDIA driver install

Once the instance has booted, verify that the NVIDIA driver has been
installed:

1. SSH to your image:

   ```
   gcloud compute ssh "DEPLOYMENT_NAME"
   ```
2. Run the following command:

   ```
   nvidia-smi
   ```

If the drivers have been installed, you can continue.

## Stop the instance

Stop the instance by running the following command:

```
gcloud compute instances stop "DEPLOYMENT_NAME"
```

## Create your own image

Now you create your own image based on the stopped instance. Run the following
at the command line, giving the new image a name and a family name:

```
gcloud compute images create "NEW_IMAGE_NAME" \
        --source-disk DEPLOYMENT_NAME \
        --source-disk-zone ZONE \
        --family NEW_FAMILY_NAME
```

Once the command is finished running, you have a new image with NVIDIA drivers
preinstalled that you can use to create new Deep Learning VM instances.