Help Center/ DataArts Fabric/ Billing/ Overview/ Ray and XDS Billing Overview
Updated on 2025-07-11 GMT+08:00
View PDF
Share

Ray and XDS Billing Overview

This document describes the billing modes, billing items, and renewal of Ray and XDS.

Billing Mode

DataArtsFabric offers two billing modes: yearly/monthly and pay-per-use.

  • Yearly/Monthly: This is a prepaid billing mode where you are charged based on the duration of your resource purchase. Longer purchase periods typically receive greater discounts. This mode is ideal for scenarios with stable compute resource demand throughout the purchase duration.
  • Pay-per-use: With pay-per-use, you are billed after consuming resources, based on the actual usage duration of compute resources. Billing is precise to the second and settled hourly. This mode provides flexible resource usage aligned with service requirements, eliminating the need for pre-provisioning and preventing resource over- or under-allocation. It is suitable for scenarios where resource requirements fluctuate, ensuring you only pay for what you use.
    Table 1 DataArtsFabric billing modes

    Billing Mode

    Payment Method

    Billing Cycle

    Applicable Billing Item

    Yearly/Monthly

    Prepaid.

    Billed by the purchase period specified in your order.

    Billed by the purchase period specified in the order

    Ray resources

    Pay-per-use

    Postpaid.

    Billed by usage duration.

    Billed based on the actual usage duration of resources. A bill is generated every hour.

    Ray resources, MU hours, SQL warm-up resources, and SQL compute unit hours

The following figure shows the billing modes of DataArtsFabric in different service scenarios.

Billing Items

DataArtsFabric implements distinct billing policies based on its Ray and inference service scenarios.

Table 2 Billing items

Billing Item

Billing description

Ray resources

You are billed based on the specifications and quantity of Ray resources provisioned. Pricing varies by Data Processing Unit (DPU) or AI Compute Unit (ACU) specifications. Both yearly/monthly and pay-per-use billing modes are available.

Model compute unit hours

Billing is based on the compute unit hours consumed by model instances deployed on inference endpoints. This item supports pay-per-use billing. The cost is calculated as: (Number of model instances under an inference endpoint) × (Number of compute units) × (Usage duration reported in seconds). Refer to Common Models for specific compute unit requirements of different base models.

Bills

On the top navigation bar of the console, click Billing. In the left navigation pane of the Billing Center console, choose Billing > Transaction and Detailed Bills to view bill details. For details, see Bills.

Renewal/Arrears

For details, see Expiration and Renewal.

Stopping Billing

If you no longer need to use your cloud services, you can unsubscribe from or delete them to stop the billing. For details, see Stopping Billing.

Billing Examples

The following prices are for reference only. For the actual prices, see pricing details for each service.

  1. DataArtsFabric Ray resources

    DataArtsFabric Ray resources are billed on a pay-per-use basis, with charges calculated per second. The hourly price is detailed in the DataArtsFabric product price list; divide this by 3600 to determine the per-second rate.

    Example: If a fabric.ray.dpu.d1x pay-per-use instance costs ¥0.2/hour, and you use 5 instances:

    • For 30 minutes of usage: (0.2/3600) x 5 x 30 x 60 = ¥0.5.
    • For 1 hour of usage: (0.2/3600) x 5 x 60 x 60 = ¥1.
  2. DataArtsFabric model computing unit (MU) hours

    MU hours are billed on a pay-per-use basis, with charges calculated per second. The hourly price is specified in the DataArtsFabric product price details; divide this by 3600 to obtain the per-second rate.

    Example: A base model, such as LLAMA3_8B, consumes 2 MUs per inference endpoint instance. Assuming the MU hour price is ¥30.0/hour, the price is calculated based on the number of deployed model instances, multiplied by the MU consumption per instance, and then by the actual usage duration (seconds).

    • If one model service instance is used for 30 minutes (and the number of instances remains constant), the price is: (1/3600) x 1 x 2 x 30 x 60 = ¥30.
    • If, over 1 hour, two service instances are deployed for the first 15 minutes, and then one service instance for the remaining 45 minutes, the price is: (30/3600) x 1 x 2 x 45 x 60 + (30/3600) x 2 x 2 x 15 x 60 = ¥75.
Parent topic: Overview