Use different default image when using with gpu instances

[sansmoraxz]

The default ubuntu images lacks a lot of stuff when trying to create with GPU enabled.

Currently I am using Deep Learning Base GPU AMI (Ubuntu 20.04) 20231026 ie ami-0e376370bca957974 for region ap-south-1 to create GPU based dev-containers. It would be ideal if it was inferred from requested devcontainer config file, as we have to manually look up the ami while creating for other regions.

[pascalbreuninger]

Hi @sansmoraxz, thanks for opening the issue. You only have to do this when setting it up initially, how big of an issue is this for you?
I could see a possible devcontainer.json customization for this kind of stuff but generally speaking the infrastructure from the providers are mostly separated from the devcontainer spec

[sansmoraxz]

Not huge, more of a QOL thing for me. I do have seperate provides for use cases. And have a bash script to update the amis.

But someone with less understanding may struggle here, or even waste time trying to debug the issue.

Related: loft-sh/devpod#669

[sansmoraxz]

As I understand a condition for whether GPU is enabled can be passed here:

devpod-provider-aws/pkg/aws/aws.go

Lines 176 to 192 in b8c145e

	Name: aws.String("root-device-type"),
	Values: []string{
	"ebs",
	},
	},
	{
	Name: aws.String("platform-details"),
	Values: []string{
	"Linux/UNIX",
	},
	},
	{
	Name: aws.String("description"),
	Values: []string{
	"Canonical, Ubuntu, 22.04 LTS*",
	},
	},

Would've made a PR, but I didn't get how much API changes needed to be made to get it working.