This topic describes how Elastic Algorithm Service (EAS) of Machine Learning Platform for AI (PAI) is billed.

You are charged EAS fees for the resources that are used to deploy model services.
  • You can deploy model services in shared resource groups and dedicated resource groups. Bills are generated based on the used resources of resource groups. For more information about the differences between the two types of resource groups, see Dedicated resource groups.
  • If you deploy a model service in a shared resource group and a dedicated resource group, you are charged for using the resources of both resource groups.
The following table describes the billing rules for shared resource groups and dedicated resource groups.
Resource Billable item Billing method Billing rule Method to stop billing
Shared resource group The amount of time for which a model service has been running. The value equals the amount of time for which the model services have occupied the shared resources. Pay-as-you-go Bills are generated based on the amount of time for which the model services have occupied shared resources. The billing starts immediately after model services are created. Stop model services
Dedicated resource group The amount of time for which the resource groups have been running. Pay-as-you-go Only dedicated resource groups are billed. Model services deployed in dedicated resource groups are not charged. The billing starts immediately after pay-as-you-go dedicated resource groups are created. Stop dedicated resource groups
Subscription N/A

Billing rules for shared resource groups

You can use the default instance type or specify the instance type as needed if you use shared resource groups. Billing rules vary based on the instance type that you use.
Item Default instance type Specified instance type
Billing formula
Bill amount of each model service = Number of instances × (Number of vCPUs × Unit price of vCPU + Memory size × Unit price of memory) × Usage duration
Note Usage duration is measured in minutes.
Bill amount of each model service = Number of instances × Unit price × Usage duration
Note Usage duration is measured in minutes.
Unit price For more information about the prices, see Prices of shared resource groups with the default instance type. For more information about the prices, see Prices of shared resource groups with the specified instance type.
Usage duration
  • Bills are generated from the point in time when a model service starts to run and consume resources.
  • The system stops billing immediately after a model service is stopped and releases resources.
Scaling
  • If you scale out a model service, newly added resources are billed after the scale-out activity is complete.
  • If you scale in a model service, released resources are no longer billed, and only the remaining resources are billed.
Notes
  • Fees are charged on a per-minute basis. No fee is charged if a resource group is occupied for less than one minute.
  • To prevent unnecessary costs, we recommend that you stop unwanted model services.
  • Fees are charged on a per-minute basis. No fee is charged if a resource group is occupied for less than one minute.
  • To prevent unnecessary costs, we recommend that you stop unwanted model services.
  • Some resources may be unavailable in one or more regions for a short period of time. During the time period, you cannot purchase relevant resource groups in these regions.

Prices of shared resource groups with the default instance type

If you use pay-as-you-go shared resource groups with the default instance type, fees are charged on a per-minute basis. The following table describes the unit prices per hour. You can calculate the unit prices per minute by dividing the prices listed in the table by 60.
Resource type Unit price Region
CPU USD 0.03 per vCPU-hour
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
  • China (Beijing)
  • China (Hong Kong)
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • India (Mumbai)
  • Germany (Frankfurt)
  • US (Virginia)
Memory USD 0.004 per GB-hour

Prices of shared resource groups with the specified instance type

The pay-as-you-go billing method is used for the shared resource groups with the specified instance type. The prices vary with instance types and regions. The following table describes the unit prices per hour. You can calculate the unit prices per minute by dividing the prices listed in the table by 60.
Instance type Instance family Number of vCPUs Memory (GiB) Unit price (USD per hour) Region
ecs.c5.6xlarge c5, compute-optimized instance family 24 48 1.08
  • India (Mumbai)
  • US (Virginia)
1.26
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
  • Singapore (Singapore)
1.2
  • Indonesia (Jakarta)
  • China (Hong Kong)
0.96 China (Zhangjiakou)
1.14 Germany (Frankfurt)
ecs.c6.2xlarge c6, compute-optimized instance family 8 16 0.36
  • India (Mumbai)
  • US (Virginia)
  • US (Silicon Valley)
0.42
  • Indonesia (Jakarta)
  • Singapore (Singapore)
  • Germany (Frankfurt)
0.24
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.48 China (Hong Kong)
0.18 China (Zhangjiakou)
ecs.c6.4xlarge c6, compute-optimized instance family 16 32 0.72
  • India (Mumbai)
  • US (Virginia)
  • US (Silicon Valley)
0.84
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
0.54
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.9 China (Hong Kong)
0.36 China (Zhangjiakou)
ecs.c6.6xlarge c6, compute-optimized instance family 24 48 1.08
  • India (Mumbai)
  • US (Virginia)
  • US (Silicon Valley)
1.26
  • Singapore (Singapore)
  • Indonesia (Jakarta)
0.78
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.38 China (Hong Kong)
0.54 China (Zhangjiakou)
1.2 Germany (Frankfurt)
ecs.c6.8xlarge c6, compute-optimized instance family 32 64 1.44
  • India (Mumbai)
  • US (Virginia)
  • US (Silicon Valley)
1.68
  • Singapore (Singapore)
  • Indonesia (Jakarta)
1.08
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.8 China (Hong Kong)
0.72 China (Zhangjiakou)
1.62 Germany (Frankfurt)
ecs.g5.6xlarge g5, general-purpose instance family 24 96 1.8
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.38
  • China (Zhangjiakou)
  • US (Virginia)
1.68 China (Hong Kong)
1.32 India (Mumbai)
1.74 Singapore (Singapore)
1.62 Indonesia (Jakarta)
1.56 Germany (Frankfurt)
ecs.g6.2xlarge g6, general-purpose instance family 8 32 0.48 India (Mumbai)
0.54
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
0.36
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.6 China (Hong Kong)
0.24 China (Zhangjiakou)
0.42
  • US (Virginia)
  • US (Silicon Valley)
ecs.g6.4xlarge g6, general-purpose instance family 16 64 0.9
  • India (Mumbai)
  • US (Virginia)
  • US (Silicon Valley)
1.08
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
0.66
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.2 China (Hong Kong)
0.48 China (Zhangjiakou)
ecs.g6.6xlarge g6, general-purpose instance family 24 96 1.38 India (Mumbai)
1.68
  • Singapore (Singapore)
  • Indonesia (Jakarta)
1.02
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.8 China (Hong Kong)
0.72 China (Zhangjiakou)
1.62 Germany (Frankfurt)
1.32
  • US (Virginia)
  • US (Silicon Valley)
ecs.g6.8xlarge g6, general-purpose instance family 32 128 1.86 India (Mumbai)
2.22
  • Singapore (Singapore)
  • Indonesia (Jakarta)
1.38
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
2.4 China (Hong Kong)
0.96 China (Zhangjiakou)
1.8
  • US (Virginia)
  • US (Silicon Valley)
ecs.gn5-c28g1.7xlarge gn5, GPU-accelerated compute-optimized instance family (NVIDIA P100 × 1) 28 112 3.78 India (Mumbai)
3.72
  • Singapore (Singapore)
  • Indonesia (Jakarta)
4.08
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
3.54 China (Hong Kong)
3.84 Germany (Frankfurt)
3.48
  • US (Virginia)
  • US (Silicon Valley)
ecs.gn5-c4g1.xlarge gn5, GPU-accelerated compute-optimized instance family (NVIDIA P100 × 1) 4 30 2.04
  • India (Mumbai)
  • China (Hong Kong)
  • US (Silicon Valley)
2.16
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.98 China (Zhangjiakou)
1.92 Germany (Frankfurt)
1.86 US (Virginia)
ecs.gn5-c8g1.2xlarge gn5, GPU-accelerated compute-optimized instance family (NVIDIA P100 × 1) 8 60 2.46
  • India (Mumbai)
  • China (Hong Kong)
  • US (Silicon Valley)
2.58
  • Singapore (Singapore)
  • Indonesia (Jakarta)
2.64
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
2.34 China (Zhangjiakou)
2.28 Germany (Frankfurt)
2.22 US (Virginia)
ecs.gn5-c8g1.4xlarge gn5, GPU-accelerated compute-optimized instance family (NVIDIA P100 × 2) 16 120 4.92
  • China (Hong Kong)
  • India (Mumbai)
5.16
  • Singapore (Singapore)
  • Indonesia (Jakarta)
5.22
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
4.74 China (Zhangjiakou)
4.62 Germany (Frankfurt)
4.44 US (Virginia)
4.86 US (Silicon Valley)
ecs.gn5i-c4g1.xlarge gn5i, GPU-accelerated compute-optimized instance family (NVIDIA P4 × 1) 4 16 1.68
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.26 China (Zhangjiakou)
ecs.gn5i-c8g1.2xlarge gn5i, GPU-accelerated compute-optimized instance family (NVIDIA P4 × 1) 8 32 1.98
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.5 China (Zhangjiakou)
ecs.gn6i-c16g1.4xlarge gn6i, GPU-accelerated compute-optimized instance family (NVIDIA T4 × 1) 16 62 2.28
  • India (Mumbai)
  • Germany (Frankfurt)
2.34 Singapore (Singapore)
2.22 Indonesia (Jakarta)
2.82
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
2.1 China (Hong Kong)
2.52 China (Zhangjiakou)
1.98 US (Virginia)
ecs.gn6i-c24g1.12xlarge gn6i, GPU-accelerated compute-optimized instance family (NVIDIA T4 × 2) 48 186 5.46 India (Mumbai)
5.94 Singapore (Singapore)
5.52 Indonesia (Jakarta)
5.88
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
5.22 China (Hong Kong)
5.28 China (Zhangjiakou)
5.7 Germany (Frankfurt)
5.04 US (Virginia)
ecs.gn6i-c24g1.6xlarge gn6i, GPU-accelerated compute-optimized instance family (NVIDIA T4 × 1) 24 93 2.7 India (Mumbai)
2.76 Indonesia (Jakarta)
2.94
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
  • Singapore (Singapore)
2.64
  • China (Hong Kong)
  • China (Zhangjiakou)
2.88 Germany (Frankfurt)
2.52 US (Virginia)
ecs.gn6i-c4g1.xlarge gn6i, GPU-accelerated compute-optimized instance family (NVIDIA T4 × 1) 4 15 1.5
  • India (Mumbai)
  • Singapore (Singapore)
1.44 Indonesia (Jakarta)
1.98
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.32
  • China (Hong Kong)
  • US (Virginia)
1.8 China (Zhangjiakou)
1.38 Germany (Frankfurt)
ecs.gn6i-c8g1.2xlarge gn6i, GPU-accelerated compute-optimized instance family (NVIDIA T4 × 1) 8 31 1.74 India (Mumbai)
1.8 Singapore (Singapore)
1.68
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
2.4
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.56
  • China (Hong Kong)
  • US (Virginia)
2.16 China (Zhangjiakou)
ecs.gn6v-c8g1.2xlarge gn6v, GPU-accelerated compute-optimized instance family (NVIDIA V100 × 1) 8 32 5.22 Singapore (Singapore)
4.5
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
3.36 China (Zhangjiakou)
3.12 US (Virginia)
ecs.r6.2xlarge r6, memory-optimized instance family 8 64 0.6
  • India (Mumbai)
  • US (Virginia)
  • US (Silicon Valley)
0.72
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
0.48
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.78 China (Hong Kong)
0.3 China (Zhangjiakou)
ecs.r6.4xlarge r6, memory-optimized instance family 16 128 1.2 India (Mumbai)
1.38
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
0.9
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.56 China (Hong Kong)
0.6 China (Zhangjiakou)
1.14
  • US (Virginia)
  • US (Silicon Valley)
ecs.r6.6xlarge r6, memory-optimized instance family 24 192 1.8 India (Mumbai)
1.38
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
2.1
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
2.34 China (Hong Kong)
0.96 China (Zhangjiakou)
1.74
  • US (Virginia)
  • US (Silicon Valley)
ecs.r6.8xlarge r6, memory-optimized instance family 32 256 2.4 India (Mumbai)
2.76
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
1.8
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
3.06 China (Hong Kong)
1.26 China (Zhangjiakou)
2.34
  • US (Virginia)
  • US (Silicon Valley)
ecs.g7.2xlarge g7, general-purpose instance family 8 32 0.54
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • China (Hong Kong)
0.42
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
ecs.g7.4xlarge g7, general-purpose instance family 16 64 1.02
  • Singapore (Singapore)
  • Indonesia (Jakarta)
0.78
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.14 China (Hong Kong)
ecs.g7.6xlarge g7, general-purpose instance family 24 96 1.56
  • Singapore (Singapore)
  • Indonesia (Jakarta)
1.2
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.68 China (Hong Kong)
ecs.g7.8xlarge g7, general-purpose instance family 32 128 2.04
  • Singapore (Singapore)
  • Indonesia (Jakarta)
1.56
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
2.28 China (Hong Kong)
ecs.c7.2xlarge c7, compute-optimized instance family 8 16 0.42
  • Singapore (Singapore)
  • Indonesia (Jakarta)
0.3
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.48 China (Hong Kong)
ecs.c7.4xlarge c7, compute-optimized instance family 16 32 0.84
  • Singapore (Singapore)
  • Indonesia (Jakarta)
0.6
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.9 China (Hong Kong)
ecs.c7.6xlarge c7, compute-optimized instance family 24 48 1.26
  • Singapore (Singapore)
  • Indonesia (Jakarta)
0.9
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.38 China (Hong Kong)
ecs.c7.8xlarge c7, compute-optimized instance family 32 64 1.68
  • Singapore (Singapore)
  • Indonesia (Jakarta)
1.2
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.86 China (Hong Kong)
ecs.r7.2xlarge r7, memory-optimized instance family 8 64 0.66 Singapore (Singapore)
0.54
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.72 China (Hong Kong)
ecs.r7.4xlarge r7, memory-optimized instance family 16 128 1.32 Singapore (Singapore)
1.02
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
1.44 China (Hong Kong)
ecs.r7.6xlarge r7, memory-optimized instance family 24 192 1.98 Singapore (Singapore)
1.56
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
2.16 China (Hong Kong)
ecs.r7.8xlarge r7, memory-optimized instance family 32 256 2.58 Singapore (Singapore)
2.04
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
2.88 China (Hong Kong)
ecs.gn7-c12g1.3xlarge gn7, GPU-accelerated compute-optimized instance family (NVIDIA A100 × 1) 12 95 5.4
  • China (Beijing)
  • China (Shanghai)
ecs.g7.16xlarge g7, general-purpose instance family 64 256 4.08
  • Singapore (Singapore)
  • Indonesia (Jakarta)
3.12
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
4.5 China (Hong Kong)
ecs.c7.16xlarge c7, compute-optimized instance family 64 128 3.36
  • Singapore (Singapore)
  • Indonesia (Jakarta)
2.46
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
3.66 China (Hong Kong)
ecs.r7.16xlarge r7, memory-optimized instance family 64 512 5.22 Singapore (Singapore)
4.08
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
5.7 China (Hong Kong)
ecs.gn7i-c8g1.2xlarge gn7i, GPU-accelerated compute-optimized instance family (NVIDIA A10 × 1) 8 30 3.06 Singapore (Singapore)
2.16
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
ecs.gn7i-c16g1.4xlarge gn7i, GPU-accelerated compute-optimized instance family (NVIDIA A10 × 1) 16 60 3.24 Singapore (Singapore)
2.28
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
ecs.gn7i-c32g1.8xlarge gn7i, GPU-accelerated compute-optimized instance family (NVIDIA A10 × 1) 32 188 3.6 Singapore (Singapore)
2.52
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
ecs.gn6e-c12g1.3xlarge gn6e, GPU-accelerated compute-optimized instance family (NVIDIA V100 × 1) 12 92 4.68 Singapore (Singapore)
3.36
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
ecs.g6.xlarge g6, general-purpose instance family 4 16 0.24
  • India (Mumbai)
  • Germany (Frankfurt)
  • US (Virginia)
  • US (Silicon Valley)
0.3
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • China (Hong Kong)
0.18
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.12 China (Zhangjiakou)
ecs.c6.xlarge c6, compute-optimized instance family 4 8 0.18
  • India (Mumbai)
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • Germany (Frankfurt)
  • US (Virginia)
  • US (Silicon Valley)
0.12
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
  • China (Zhangjiakou)
0.24 China (Hong Kong)
ecs.r6.xlarge r6, memory-optimized instance family 4 32 0.3
  • India (Mumbai)
  • US (Virginia)
  • US (Silicon Valley)
0.36
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • China (Hong Kong)
  • Germany (Frankfurt)
0.24
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
0.18 China (Zhangjiakou)
ecs.g6.large g6, general-purpose instance family 2 8 0.12
  • India (Mumbai)
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • China (Hong Kong)
  • Germany (Frankfurt)
  • US (Virginia)
  • US (Silicon Valley)
0.06
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
  • China (Zhangjiakou)
ecs.c6.large c6, compute-optimized instance family 2 4 0.12
  • India (Mumbai)
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • China (Hong Kong)
  • Germany (Frankfurt)
  • US (Virginia)
  • US (Silicon Valley)
0.06
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
  • China (Zhangjiakou)
ecs.r6.large r6, memory-optimized instance family 2 16 0.12
  • India (Mumbai)
  • China (Beijing)
  • China (Hangzhou)
  • China (Shanghai)
  • China (Shenzhen)
  • US (Virginia)
  • US (Silicon Valley)
0.18
  • Singapore (Singapore)
  • Indonesia (Jakarta)
  • China (Hong Kong)
  • Germany (Frankfurt)
0.06 China (Zhangjiakou)

Billing examples for shared resource groups

  • Scenario:
    You used shared resource groups with the specified instance type in the China (Hangzhou) region to deploy a model service.
    • The model service occupied 2 vCPUs and 8 GB of memory, and started to run at 09:00:00 (UTC+8) on June 3, 2019.
    • You scaled in the model service and reduced the occupied resources to 1 vCPU and 4 GB of memory at 10:00:00 (UTC+8) on June 3, 2019.
    • Then, you scaled out the model service and increased the occupied resources to 4 vCPUs and 16 GB of memory at 11:00:00 (UTC+8) on June 3, 2019.
    • At 12:00:00 (UTC+8) on June 3, 2019, the model service stopped running.
  • The bill amount is calculated based on the following formula:
    Bill amount: 2 × 0.03 + 8 × 0.004 + 1 × 0.03 + 4 × 0.004 + 4 × 0.03 + 16 × 0.004 = USD 0.322

Billing rules for dedicated resource groups

Two types of billing methods are supported for dedicated resource groups: subscription and pay-as-you-go. Billing rules vary based on the two types of billing methods.
Item Subscription Pay-as-you-go
Billing formula
Bill amount of each resource group = Number of resources × Unit price × Subscription duration
Note Subscription duration is measured in minutes.
Bill amount of each resource group = Number of resources × (Unit price/60) × Usage duration
Note Subscription duration is measured in minutes.
Unit price For more information about the prices, see Prices of dedicated resource groups.
Subscription or usage duration After you purchase a dedicated resource group, the resource group is free of charge on the current day, and the subscription takes effect from the next day. For example, you purchased a dedicated resource group on July 31, 2019, and the subscription duration is one month. Then, the resource group expired at 00:00:00 (UTC+8) on August 31, 2019.
Note Valid subscription durations: one month, two months, three months, and six months.
  • The billing starts immediately after a dedicated resource group is created and enters the Running state.
  • The billing stops immediately after the resource group enters the No Node state.
Scaling N/A
  • If you scale out a model service, newly added resources are billed after the scale-out activity is complete.
  • If you scale in a model service, released resources are no longer billed, and only the remaining resources are billed.
Notes Some resources may be unavailable in one or more regions for a short period of time. During the time period, you cannot purchase relevant resource groups in these regions.
  • Fees are charged on a per-minute basis. No fee is charged if a resource group is occupied for less than one minute.
  • To prevent unnecessary costs, we recommend that you stop unwanted model services.
  • Some resources may be unavailable in one or more regions for a short period of time. During the time period, you cannot purchase relevant resource groups in these regions.

Prices of dedicated resource groups

For more information about the pricing of the subscription dedicated resource groups, go to the Dedicated Node for EAS (Subscription) buy page.

For more information about the pricing of pay-as-you-go dedicated resource groups, go to the Dedicated Node for EAS (Pay-As-You-Go) buy page.

Billing examples for dedicated resource groups

  • Subscription
    • Scenario:

      You subscribe to two subscription NVIDIA T4 GPUs in the China (Hangzhou) region for three months, and the specifications of each GPU are 4 vCPUs and 15 GB of memory. The unit price of the subscribed GPUs is USD 570 per month.

    • The bill amount is calculated based on the following formula:
      Bill amount: 2 × 570 × 3 = USD 3,420
  • Pay-as-you-go
    • Scenario:

      You purchase two pay-as-you-go Elastic Compute Service (ECS) instances whose instance type is ecs.g6.6xlarge in the China (Hangzhou) region, and use the instances for 45 minutes. The specifications of each ECS instance are 24 vCPUs and 96 GB of memory. The unit price of the ECS instance of the ecs.g6.6xlarge type is USD 2.94 per hour.

    • The bill amount is calculated based on the following formula:
      Bill amount: 2 × (2.94/60) × 45 = USD 4.41