I am trying to create a service in my ECS cluster to run a task that uses elastic inference accelerator (EIA). However, when I try to create the service, I get the following error:
I have read the docs, and I understand there cannot be more than 5 accelerators in the same EC2 intance, and thus I have created a brand new EC2 for this.
But I still have the issue. If i remove the EI accelerator from the task definition, everything works fine. My task defintion reference looks like this:
"resourceRequirements": [
{
"type": "InferenceAccelerator",
"value": "device-gpu2"
}
And in the container definition:
"eia": "device-gpu2"
PS: I have also double checked that my cloudwatch log groups have been created. I have several other EC2 instancers that work perfectly with the same config, so this is quite wierd, and I am positive it must be a limit issue, but with the wrong error message from the AWS console.
I have another EC2 instance that has 5 accelerators attached through ECS, and it works fine. Do you have any idea please?