Skip to content

ECS Managed Instance supports g6f instances?

0

I am trying to build an application with a small LLM by using a g6f.xlarge instance. I am trying to use an ECS Managed Instance. But the creation of a capacity provider is not working.

According to this doc:

https://aws.amazon.com/ec2/instance-types/g6/

4 vpcu and memory 16 gb

aws ecs create-capacity-provider \
  --name "test-g6f-cp-cli" \
  --cluster "synapse-stg" \
  --managed-instances-provider '{
    "infrastructureRoleArn": "arn:aws:iam::12345678:role/ecsInfrastructureRole",
    "instanceLaunchTemplate": {
      "ec2InstanceProfileArn": "arn:aws:iam::12345678:instance-profile/ecsInstanceProfile",
      "instanceRequirements": {
        "vCpuCount": {"min": 4, "max": 4},
        "memoryMiB": {"min": 16384, "max": 16384},
        "allowedInstanceTypes": ["g6f.xlarge"]
      },
      "networkConfiguration": {
        "subnets": ["subnet-abcdefg", "subnet-abdefgf"],
        "securityGroups": ["sg-0212131sbcsabc"]
      },
      "storageConfiguration": {
        "storageSizeGiB": 50
      },
      "monitoring": "basic"
    }
  }' \
  --region us-east-2

An error occurred (ClientException) when calling the CreateCapacityProvider operation: No instance types satisfy the instance requirements specified in the Managed Instances capacity provider.

asked 10 months ago353 views

3 Answers
0

Hello.

I think you need settings such as "AcceleratorCount" and "AcceleratorManufacturers".
The following document describes the parameters required when using a GPU.
Try including the following parameters in your "instanceRequirements":
https://docs.aws.amazon.com/AmazonECS/latest/developerguide/managed-instances-gpu.html

{
  "instanceRequirements": {
    "acceleratorTypes": "gpu",
    "acceleratorCount": { "min": 1, "max": 3 },
    "acceleratorManufacturers": ["nvidia"]
  }
}
EXPERT

answered 10 months ago

  • Same result. The issue seems to be that g6f.xlarge has a fractional GPU (1/8 of an L4), so acceleratorCount: {"min": 1} doesn't match because it doesn't have 1 whole GPU.

  • Even if you try from the management console, g6f cannot be selected, so it may not be currently available for ECS managed instances. The documentation states the following, so it seems to be supported... I'll check with AWS support. https://docs.aws.amazon.com/AmazonECS/latest/developerguide/managed-instances-instance-types.html

    g6, g6e, g6f: Latest generation GPU instances

  • After checking with AWS support, it appears that there is a bug that makes it impossible to select g6f-based instance types. They say they are currently working on a fix, but they cannot share when it will be completed.

-1

Amazon ECS Managed Instances supports a broad range of EC2 instance types, including GPU-accelerated instances, but there appears to be an issue with your specific configuration for g6f.xlarge instances.

The error message "No instance types satisfy the instance requirements specified" suggests that the combination of parameters you've specified isn't matching available instance types in your region. This could be due to several reasons:

  1. The g6f.xlarge instance type might not be available in the us-east-2 region where you're trying to create the capacity provider.

  2. Your configuration is too restrictive. You've set both vCPU count and memory to exact values (4 vCPUs and 16384 MiB), which leaves no flexibility for the service to select alternative instances if g6f.xlarge isn't available.

  3. There might be capacity constraints for g6f.xlarge instances in your selected Availability Zones.

To resolve this issue, you could try:

  • Widening your instance requirements by providing a range for vCPU and memory rather than exact values
  • Checking if g6f.xlarge is available in your region and the specific Availability Zones corresponding to your subnets
  • Adding alternative GPU instance types to your allowedInstanceTypes list
  • Using attribute-based instance selection to specify GPU requirements instead of explicitly naming instance types

When creating a custom capacity provider for GPU workloads, you can use attribute-based instance type selection to specify accelerator types and requirements, which gives ECS Managed Instances more flexibility in selecting appropriate instance types while still meeting your GPU needs.
Sources
Announcing Amazon ECS Managed Instances for containerized applications | AWS News Blog
CloudFormation example templates for Amazon ECS - Amazon Elastic Container Service
Architect for Amazon ECS Managed Instances - Amazon Elastic Container Service
Creating a capacity provider for Amazon ECS Managed Instances - Amazon Elastic Container Service

answered 10 months ago

-1

Hi,

The error you’re seeing — “No instance types satisfy the instance requirements specified in the Managed Instances capacity provider” — is not caused by a mistake in your configuration. Instead, it appears to come from how ECS Managed Instances currently recognizes (or fails to recognize) the G6f instance family, especially in the us-east-2 region.

According to AWS documentation, ECS Managed Instances does support the G6, G6e, and G6f GPU families. The specifications of g6f.xlarge — 4 vCPUs and 16 GiB of memory — also perfectly match the requirements you defined in instanceRequirements. So your JSON configuration is valid.

The issue arises from how ECS Managed Instances evaluates instance types internally. When you create a capacity provider with a Managed Instances provider, ECS does the following:

  1. Builds a list of instance types that ECS Managed Instances believes it supports.
  2. Applies your constraints (vCPU, memory, allowedInstanceTypes, etc.).
  3. If the intersection of those two lists is empty, it throws exactly the error you’re seeing.

In your case, the most likely explanation is this:

EC2 fully supports g6f.xlarge in us-east-2, but the ECS Managed Instances backend does not yet recognize this instance family as eligible, even though the public documentation says it is supported.

This results in ECS saying “no instance type matches,” even though g6f.xlarge objectively should match.

A practical way to confirm this is to test with other GPU instance types. If you try the exact same configuration but substitute g6.xlarge or g5.xlarge, the capacity provider will likely create successfully. You can also try a more relaxed requirement (for example, using only acceleratorTypes: ["gpu"] and no allowedInstanceTypes) to see whether ECS can create a GPU-backed Managed Instance at all. If this works while g6f.xlarge alone fails, then the issue is clearly isolated to G6f.

One more factor to check is whether your account can launch g6f.xlarge in the specific availability zones where your ECS subnets live. Running describe-instance-type-offerings will confirm that. However, based on current reports, the most common cause is not AZ availability but ECS Managed Instances lagging behind EC2 in recognizing new GPU instance families.

In summary, nothing is wrong with your configuration. The behavior strongly suggests that ECS Managed Instances has not fully integrated support for the G6f family in your region, despite the official documentation stating otherwise. If your tests confirm that other GPU instance types work but G6f does not, the next step is to open an AWS Support case and ask whether G6f instance types are currently enabled for ECS Managed Instances in us-east-2.

If this resolved your issue, please accept the answer to help others find this solution. thx.

answered 10 months ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.