View a markdown version of this page

When to use Amazon ECS Managed Instances - AWS Batch

When to use Amazon ECS Managed Instances

Use Amazon ECS Managed Instances when your jobs need capabilities beyond what Fargate provides and you prefer not to manage the underlying Amazon EC2 infrastructure. Amazon ECS Managed Instances combines the operational simplicity of Fargate with the compute flexibility of Amazon EC2.

Use Amazon ECS Managed Instances if your jobs require any of the following:

  • Your jobs require GPU instances (NVIDIA accelerators) for machine learning or rendering workloads.

  • Your jobs need more than 32 vCPUs or 244 GiB of memory.

  • Your containers require privileged access.

  • Your containers need host-level devices or volumes.

  • You need specific Amazon EC2 instance types or instance families.

  • You want to use On-Demand Capacity Reservations, Reserved Instances, or Amazon EC2 Instance Savings Plans.

  • You need ARM64 (Graviton) instances via runtimePlatform configuration.

  • You need arbitrary vCPU and memory combinations without the fixed size pairings that Fargate requires.

Fargate is the simplest and recommended option if you don't have specific requirements for the underlying compute infrastructure. Continue to use Fargate if:

  • Your jobs fit within Fargate resource limits (up to 32 vCPUs, 244 GiB memory).

  • You need host-level isolation (each Fargate task runs in its own dedicated kernel runtime environment).

  • You don't need GPU, privileged access, or host-level resources.

Use Amazon EC2 managed compute environments if:

  • You need multi-node parallel (MNP) jobs.

  • You require custom AMIs or launch templates.

  • You need fine-grained control over allocation strategies.

  • You need a minimum capacity warm pool (minvCpus).