How to manage GPU resources for a single worker in a distributed folder?

I have a question about dask is distributed .

Suppose I would like to run a set of tasks, each running on a different number of GPUs, for example, one task is running on two GPUs (type A), while several others are running on 1 GPU (type B). My understanding is that it is possible to configure multiple workers with the "resources" property. For example, I might have 3 GPUs in total and set up one worker with the resource variable "gpus = 2" and another with "gpus = 1". It is clear that here "gpus" is an abstract term (see here ).

However, this would mean that after a type A task is launched in a workstation with two GPUs, that worker will execute one type B job at the same time, thus not using another available GPU associated with that worker. It's right?

Ideally, I would like to define one worker with a total GPU count (here: 3) that gets updated (like ncores) when jobs run. So a worker should be able to run one type A task and one type B task at the same time, and then run three type B tasks in parallel. Is it possible?

Greetings