AWS ECS Task Error -- Insufficient GPU resource available

I got the following error on an AWS ECS cluster, with launch type EC2, when I publish a service which has a task (desired count 1) with the next GPU requirements :

    "resourceRequirements": [
            {
              "type":"GPU",
              "value": "1"
            }
    ]

Error message : service XXXXXXX was unable to place a task because no container instance met all of its requirements. The closest matching container-instance YYYYYYYYY has insufficient GPU resource available.

The EC2 instance attached to the cluster is p2.8xlarge has 8 GPUS, which should be enough for my requirements.

Any help or information about how ECS computes these requirements is welcome.

Thanks