I found in Dagster's documentation a section on customizing the resource requirements for machines orchestrated by that service, in AWS ECS and over Kubernetes:
https://docs.dagster.io/dagster-cloud/deployment/agents/amazon-ecs/configuration-reference#per-deployment-configuration
https://docs.dagster.io/deployment/guides/kubernetes/customizing-your-deployment#per-job-kubernetes-configuration
And in both I can set CPU and memory.
Can I also set the available GPU resources for tasks that require those instead? (e.g. AWS' GPU based EC2 machines / here)
It appears that the answer is negative as of version 1.4.11 / 0.20.11 (libs).
The documentation seems to specify only cpu and memory. (In both server_resouces and run_resources. I assume the first is for Dagster itself, and the latter is for the workers)
Furthermore, looking at the code parsing the run resouces only 4 fields are extracted: cpu, memory, ephemeral_storage and repository_credentials. (As of the current release)