All Products
Search

This Product

    Platform For AI:EAS deployment resources

    Document Center

    Platform For AI:EAS deployment resources

    Last Updated:Apr 15, 2026

    EAS supports three resource types: public resources, EAS resource groups, and resource quota. Compare features and choose the right type for your workload.

    Resource type comparison

    Resource type

    Use cases

    Billing

    Features

    public resources

    Testing, or services with fluctuating traffic when combined with a dedicated resource group.

    • Provisioned on demand, billed after usage.

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

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

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

    EAS resource groups

    dedicated resource group

    Workloads that require high security or dedicated resources. Also useful to reserve scarce resource types in advance.

    • Requires purchase before use.

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

    • Exclusive, dedicated compute resources with resource isolation for enhanced security.

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

    • Supports GPU slicing.

    virtual resource group

    A logical group that combines multiple resource types, such as public resources, resource quota, and dedicated resource groups.

    Billed based on the resources scheduled and used.

    • Deploy a single service across multiple resource types.

    • Configurable scheduling priorities.

    resource quota

    Only Lingjun resources are supported.

    Choose a resource type

    Choose the appropriate resource type for your use case:

    Testing and development

    • Recommendation: public resources

    • Reason: Pay-as-you-go billing with no upfront cost. Suitable for test environments with unpredictable traffic.

    • Note: Resources may be insufficient during peak hours. For more information, see What to do when public resources are insufficient.

    Production environment - Stable workloads

    • Recommendation: a dedicated resource group

    • Reason: Dedicated resources ensure stable performance for high-availability workloads.

    • Cost: Supports subscription to reduce costs.

    Production environment - Fluctuating traffic

    • Recommendation: a virtual resource group (a combination of a dedicated resource group and public resources)

    • Reason: Dedicated resources provide a baseline. Public resources handle traffic spikes for cost-efficiency.

    • Tip: Set scheduling priorities to use dedicated resources first and scale out to public resources during traffic peaks.

    Special hardware requirements

    • Recommendation: resource quota (Lingjun resources)

    • Reason: Access to specific high-performance hardware.

    • Use case: Large-scale model training and inference.

    Advanced features

    After you configure resources, use the following features to optimize resource utilization:

    • GPU slicing: Splits a single GPU's compute power and memory among multiple service instances. This improves GPU utilization and reduces deployment costs. Available for dedicated resource groups and Lingjun resources.

    • Multi-node distributed inference: Deploys a single service instance across multiple machines to overcome single-node hardware limits. This enables deployment of ultra-large models such as DeepSeek 671B.

    FAQ

    See EAS FAQ.

    Related topics