Platform For AI:EAS deployment resource overview

Resource type

Scenarios

Billing

Feature comparison

Public resources

Ideal for testing or supplementing dedicated resource groups to handle traffic spikes.

• Resources are provisioned on demand and billed after use.

• Pay-as-you-go. For more information, see Billing of EAS.

• Uses shared public compute resources. No separate purchase is required, but stable resource allocation is not guaranteed during peak hours.

• Supports CPUs and GPUs (A10, P4, P100, T4, and V100).

EAS resource groups

Dedicated resource group

Ideal for reserving scarce resource types by purchasing them in advance.

• Requires purchase in advance.

• Supports subscription and pay-as-you-go. For more information, see Billing of EAS.

• Provides exclusive, dedicated compute resources, enhancing security through resource isolation.

• Supports CPUs and GPUs (A10, P4, P100, T4, and V100).

• Supports the GPU sharing feature.

Virtual resource group

Combines multiple resource types, such as public resources, resource quotas, and dedicated EAS resources, into a single logical group.

You are billed for the actual resources scheduled and used.

• Enables you to deploy a single service across multiple resource types.

• Supports setting scheduling priorities for resources.

Resource quota

Currently, only Lingjun resources are supported.

• Resources are purchased before use.

• Subscription. For more information, see Billing of AI computing resources.

• Manages integrated resources for model training and inference to improve resource utilization.

• Supports the GPU sharing feature.