View a markdown version of this page

AWS::Deadline::Fleet AcceleratorSelection - AWS CloudFormation

This is the new CloudFormation Template Reference Guide. Please update your bookmarks and links. For help getting started with CloudFormation, see the AWS CloudFormation User Guide.

AWS::Deadline::Fleet AcceleratorSelection

Describes a specific GPU accelerator required for an Amazon Elastic Compute Cloud worker host.

Syntax

To declare this entity in your CloudFormation template, use the following syntax:

JSON

{ "Name" : String, "Runtime" : String }

YAML

Name: String Runtime: String

Properties

Name

The name of the chip used by the GPU accelerator.

The available GPU accelerators are:

  • t4 - NVIDIA T4 Tensor Core GPU (16 GiB memory)

  • a10g - NVIDIA A10G Tensor Core GPU (24 GiB memory)

  • l4 - NVIDIA L4 Tensor Core GPU (24 GiB memory)

  • l40s - NVIDIA L40S Tensor Core GPU (48 GiB memory)

  • rtx-pro-server-6000 - NVIDIA RTX PRO Server 6000 GPU (96 GiB memory)

Required: Yes

Type: String

Allowed values: t4 | a10g | l4 | l40s | rtx-pro-server-6000

Update requires: No interruption

Runtime

Specifies the runtime driver to use for the GPU accelerator. You must use the same runtime for all GPUs in a fleet.

You can choose from the following runtimes:

If you don't specify a runtime, AWS Deadline Cloud uses latest as the default. However, if you have multiple accelerators and specify latest for some and leave others blank, AWS Deadline Cloud raises an exception.

Important

Not all runtimes are compatible with all accelerator types:

  • t4 and a10g: Support all runtimes (grid:r580, grid:r570, grid:r535)

  • l4 and l40s: Only support grid:r570 and newer

  • rtx-pro-server-6000: Only supports grid:r580

All accelerators in a fleet must use the same runtime version. You cannot mix different runtime versions within a single fleet.

Note

When you specify latest, it resolves to grid:r580 for all currently supported accelerators.

Required: No

Type: String

Minimum: 1

Maximum: 100

Update requires: No interruption