All Products
Search
Document Center

PAI-EAS pricing

Last Updated: Apr 03, 2020

Overview

Before using Elastic Algorithm Service (EAS) to deploy a service, you must consider the computing resources for deploying the service.EAS of Machine Learning Platform for AI (PAI) allows you to deploy model services in public resource groups and dedicated subscription resource groups.For more information about the differences between public and dedicated subscription resource groups, see Introduction to resource groups.

The EAS billing method varies depending on the resources used for the two deployment types. For more information, see the following table.

Resource group Billing item Billing method Billing rules
Public resource group The duration for which model services have run Pay-as-you-go This billing method applies to the scenario where a model service is directly deployed in a public resource group. In this case, the pay-as-you-go billing method is used based on the quantity of public computing resources that the model service occupies and the duration of occupation.
Dedicated subscription resource group The duration for which the resource group has run Subscription This billing method applies to the scenario where a model service is deployed in a dedicated subscription resource group. Since the dedicated subscription resource group is purchased in advance, you only need to pay for the resource group at this time and the model service deployed in the resource group does not incur charges. You can purchase a dedicated subscription resource group by subscription.

If you use resources of both deployment types, the total cost of EAS is the sum of all costs. The following describes how to calculate the costs of the two deployment methods.

Billing for deployment in public resource groups

1. Basic rules

According to the billing rules described in the preceding table, when a public resource group is used for deployment, the cost is calculated based on the running duration (that is, the duration of occupying public resources) of the deployed model service. The calculation method is as follows:

Cost of each model service = Number of the deployed resources * Unit price of the deployed resources * Serving duration

  • Once a model is deployed and is running, billing starts. We recommend that you stop useless model services to avoid unnecessary costs.
  • The billing start point is the time when the model (corresponding to the occupied resources) starts to run (the status changes to Running), and the billing end point is the time when the model (corresponding to the occupied resources) stops (the status changes to Stopped).
  • After the model is scaled out, the duration of new resources is calculated from the time when the scale-out is successful. After the model is scaled in, the billing of released resources stops when the scale-in is successful, and billing continues for the remaining resources.
  • The duration is accurate to the minute. If the duration is less than one minute, no charge is incurred.

The following table lists the unit price of the deployed resources, where 1 Quota = 1 core + 4 GB memory.

Resource type Unit price Region
CPU service RMB 0.4/Quota/hour China (Shanghai), China (Beijing), Singapore (Singapore), Indonesia (Jakarta), India (Mumbai), and Germany (Frankfurt)

Note: Public resource groups do not provide GPU resources. If you need GPU resources, purchase a dedicated subscription resource group.

2. Billing example

Assume that user A deployed a model service in the China (Shanghai) region. The initial resource usage was 2 quotas (2 cores + 8 GB memory). The deployment was completed at 9:00 and the service status changed to Running. At 10:00, the user completed scale-in, and the occupied resources decreased to 1 quota (1 core + 4 GB memory). At 11:00, the user completed scale-out, and the occupied resources increased to 4 quotas (4 cores + 16 GB memory). At 12:00, the user stopped the service and the service status changed to Stopped. In this case, the cost incurred by the user is calculated as follows:

Number of the deployed resources * Unit price of the deployed resources * Serving duration = 2 (Quotas) * 0.4 (RMB) * 1 (h) + 1 (Quota) * 0.4 (RMB) * 1 (h) + 4 (Quotas) * 0.4 (RMB) * 1 (h) = RMB 2.8

Billing for deployment in dedicated subscription resource groups

To deploy a model service in a dedicated subscription resource group, you need to purchase the dedicated subscription resource group in advance. In this case, you only need to pay for the resource group and the model service deployed in the resource group does not incur charges. You can purchase dedicated subscription resource groups by subscription.

1. Basic rules

   Cost of each resource group = Number of purchased resources * Unit price of the resources * Subscription duration

  • You can choose the subscription duration from one month to 12 months.
  • The subscription duration starts from the next day, that is, the day of purchase is a free use time. 30 days from the next day is one month. For example, if a user purchases a resource group for one month on July 31, the subscription duration starts to be counted from August 1, and the resource group will expire at 00:00 on August 31.
  • The following table lists the unit prices of deployed resources. Where “Regions in mainland China” include: China (Shanghai), China (Beijing). Some resources may be temporarily unavailable in certain regions.


Server model


GPU configuration


CPU configuration
Unit price in different regions (per server per month)

Regions in mainland China

Singapore (Singapore)

Indonesia (Jakarta)

 

India (Mumbai)

Germany (Frankfurt)

ecs.c5.6xlarge / 24 cores, 48 GB RMB 2,360 RMB 3,850 RMB 3,677 RMB 3,177 RMB 3,458
ecs.g5.6xlarge / 24 cores, 96 GB RMB 3,200 RMB 4,810 RMB 4,811 RMB 3,824 RMB 5,077
ecs.gn5i-c4g1.xlarge 1 NVIDIA Tesla P4 4 cores, 16 GB RMB 2,920

/

/

/

/

ecs.gn5i-c8g1.2xlarge 1 NVIDIA Tesla P4 8 cores, 32 GB RMB 3,510

/

/

/

/

ecs.gn6i-c4g1.xlarge 1 NVIDIA Tesla T4 4 cores, 15 GB RMB 3,683 RMB 4,697 RMB 4,697

/

RMB 4,666
ecs.gn6i-c8g1.2xlarge 1 NVIDIA Tesla T4 8 cores, 31 GB RMB 4,435 RMB 5,570 RMB 5,570

/

RMB 5,623
ecs.gn6i-c16g1.4xlarge 1 NVIDIA Tesla T4 16 cores, 62 GB RMB 5,198 RMB 7,317 RMB 7,317

/

RMB 7,535
ecs.gn6i-c24g1.6xlarge 1 NVIDIA Tesla T4 24 cores, 93 GB RMB 5,445 RMB 9,172 RMB 9,172

/

RMB 9,511
ecs.gn5-c4g1.xlarge 1 NVIDIA P100 4 cores, 30 GB RMB 4,049 RMB 6,646 RMB 6,314 RMB 5,999 RMB 6,502
ecs.gn5-c8g1.2xlarge 1 NVIDIA P100 8 cores, 60 GB RMB 4,876 RMB 8,004 RMB 7,603 RMB 7,223 RMB 7,829
ecs.gn5-c28g1.7xlarge 1 NVIDIA P100 28 cores, 112 GB RMB 7,565 RMB 11,517 RMB 10,941 RMB 11,206 RMB 11,267
ecs.gn6v-c8g1.2xlarge 1 NVIDIA V100 8 cores, 32 GB RMB 8,382

/

/

/

/

2. Billing example

Assume that user A purchases two 4-core, 15 GB GPU T4 cards in the China (Shanghai) region for three months. The purchase cost is calculated as follows:

Number of purchased resources * Unit price of the purchased resources * Subscription duration = 2 * RMB 3,683 * 3 months = RMB 22,098