I'm sorry for the obvious, but take a look at the link below.

https://docs.aws.amazon.com/batch/latest/userguide/gpu-jobs.html

The resourceRequirements parameter for the job definition specifies the number of GPUs to be pinned to the container. This number of GPUs isn't available to any other job that runs on that instance for the duration of that job. All instance types in a compute environment that run GPU jobs must be from the p2, p3, p4, p5, g3, g3s, g4, or g5 instance families. If this isn't done a GPU job might get stuck in the RUNNABLE status.