This topic describes the billing of Deep Learning Containers (DLC).
Billing
The following figure shows the DLC billing resources.
Billing methods
The following table describes how clusters of the self-managed and public resource group types are billed.
Billable resource | Billable item | Billing rule | Billing method | How to stop billing |
Cluster of the self-managed resource group type | Resources, networks, and storage services associated with the ACK cluster | For more information, see Billing. | For more information, see Billing. | Stop the training job or wait until the job is complete. |
Cluster of the public resource group type | Running duration of the job | A job is billed based on the amount of time during which public resources are occupied by the job. | Pay-as-you-go | Stop the training job or wait until the job is complete. |
Cluster of the public resource group type
The public resource group supports only the pay-as-you-go billing method. The following table describes the billing details.
Billing method | Calculation formula | Unit price | Billing duration | Scaling description | Usage notes |
Pay-as-you-go | | For more information, see Appendix: Pricing details of the public resource group. | Running duration of the job | N/A | None |
Billing examples
The following examples are provided for reference only. The prices listed on the DLC buy page of the Machine Learning Platform for AI (PAI) console shall prevail.
Public resource group
Scenario:
Assume that you use a node of the ecs.g6.2xlarge instance type that is deployed in the China (Shanghai) region to create a training job. The job runs for 1 minute and 15 seconds.
Fee calculation:
1 × 0.6/60 × 1.25 = 0.0125. Unit: USD.
Overdue payments
Causes
You have insufficient balance in your Alibaba Cloud account.- Subscription: Your account balance is insufficient to pay for subscription fees.
- Pay-as-you-go: Your account balance is less than the bill amount of the previous billing cycle. If the system fails to deduct fees, your account enters the overdue state.
Service suspension
- In PAI, bills are generated within 4 hours after the previous billing cycle ends and displayed by using the T + 1 settlement method. If your account enters the overdue state, you are notified by text message.
- After your account enters the overdue state, your PAI resources remain available for 24 hours. After 24 hours, your resources are stopped.
View the overdue amount
Log on to the Billing Management console.
On the Account Overview page, view the outstanding amount due.
Refund policy
Pay-as-you-go fees cannot be refunded.
Subscription fees can be refunded based on the following rules:
Five-day money-back guarantee: You are eligible for a full refund for unused resources that are purchased within five days.
Partial refund: You are eligible for a refund for the remaining duration. The refund amount is calculated by using the following formula:
Refund amount = Amount paid - Amount consumed.
Refund for renewal orders: You can cancel renewal orders that have not taken effect.
Appendix: Pricing details of the public resource group
The following table lists the pricing details.
Node type | Specifications | GPU type | Unit price (USD per hour) | Region |
ecs.g6.2xlarge | 8 vCPUs + 32 GB memory | None | 0.6 |
|
ecs.g6.4xlarge | 16 vCPUs + 64 GB memory | None | 0.6 |
|
1.2 |
| |||
ecs.g6.8xlarge | 32 vCPUs + 128 GB memory | None | 1.2 |
|
1.8 |
| |||
2.4 | China (Hong Kong) | |||
ecs.gn5-c28g1.7xlarge | 28 vCPUs + 112 GB memory | 1 * NVIDIA P100 | 3.6 |
|
3 | China (Hong Kong) | |||
ecs.gn5-c4g1.xlarge | 4 vCPUs + 30 GB memory | 1 * NVIDIA P100 | 1.8 |
|
ecs.gn5-c8g1.2xlarge | 8 vCPUs + 60 GB memory | 1 * NVIDIA P100 | 2.4 |
|
ecs.gn5-c8g1.4xlarge | 16 vCPUs + 120 GB memory | 2 * NVIDIA P100 | 4.8 |
|
4.2 | Germany (Frankfurt) | |||
ecs.gn6e-c12g1.12xlarge | 48 vCPUs + 368 GB memory | 4 * NVIDIA V100 | 17.4 | India (Mumbai) |
18 | Singapore | |||
18.6 |
| |||
12.6 |
| |||
15.6 | Germany (Frankfurt) | |||
ecs.gn6e-c12g1.3xlarge | 12 vCPUs + 92 GB memory | 1 * NVIDIA V100 | 4.2 |
|
4.8 |
| |||
3 |
| |||
3.6 | Germany (Frankfurt) | |||
ecs.gn6e-c12g1.24xlarge | 96 vCPUs + 736 GB memory | 8 * NVIDIA V100 | 35.4 | India (Mumbai) |
36 | Singapore | |||
37.8 |
| |||
25.8 |
| |||
31.8 | Germany (Frankfurt) | |||
ecs.gn6v-c8g1.2xlarge | 8 vCPUs + 32 GB memory | 1 * NVIDIA V100 | 4.2 |
|
4.8 | Singapore | |||
ecs.gn6v-c8g1.8xlarge | 32 vCPUs + 128 GB memory | 4 * NVIDIA V100 | 16.8 |
|
19.8 | Singapore | |||
ecs.gn6v-c8g1.16xlarge | 64 vCPUs + 256 GB memory | 8 * NVIDIA V100 | 34.2 |
|
39.6 | Singapore | |||
ecs.gn6v-c10g1.20xlarge | 82 vCPUs + 336 GB memory | 8 * NVIDIA V100 | 35.4 |
|
ecs.r7.16xlarge | 64 vCPUs + 512 GB memory | None | 4.8 |
|
1.8 |
| |||
5.4 | China (Hong Kong) | |||
ecs.r7.xlarge | 4 vCPUs + 32 GB memory | None | 0.6 | China (Hong Kong) |
ecs.r7.2xlarge | 8 vCPUs + 64 GB memory | None | 0.6 |
|
ecs.r7.4xlarge | 16 vCPUs + 128 GB memory | None | 0.6 |
|
1.2 |
| |||
ecs.r7.6xlarge | 24 vCPUs + 192 GB memory | None | 1.8 |
|
1.2 |
| |||
ecs.r7.8xlarge | 32 vCPUs + 256 GB memory | None | 2.4 |
|
1.8 |
| |||
ecs.g7.2xlarge | 8 vCPUs + 32 GB memory | None | 0.6 |
|
ecs.g7.3xlarge | 12 vCPUs + 48 GB memory | |||
ecs.g5.2xlarge | 8 vCPUs + 32 GB memory | None | 0.6 |
|
ecs.g6.3xlarge | 12 vCPUs + 48 GB memory | |||
ecs.g7.4xlarge | 16 vCPUs + 64 GB memory | None | 1.2 |
|
0.6 |
| |||
ecs.r7.3xlarge | 12 vCPUs + 96 GB memory | None | 0.6 |
|
1.2 | China (Hong Kong) | |||
ecs.c6e.8xlarge | 32 vCPUs + 64 GB memory | None | 1.8 |
|
1.2 |
| |||
ecs.g6.6xlarge | 24 vCPUs + 96 GB memory | None | 1.2 |
|
1.8 | China (Hong Kong) | |||
ecs.g7.6xlarge | 24 vCPUs + 96 GB memory | None | 1.2 |
|
ecs.g5.4xlarge | 16 vCPUs + 64 GB memory | None | 0.6 | India (Mumbai) |
1.2 |
| |||
ecs.hfc6.8xlarge | 32 vCPUs + 64 GB memory | None | 1.8 |
|
1.2 |
| |||
ecs.g7.8xlarge | 32 vCPUs + 128 GB memory | None | 1.8 |
|
1.2 |
| |||
2.4 | China (Hong Kong) | |||
ecs.hfc6.10xlarge | 40 vCPUs + 96 GB memory | None | 2.4 |
|
1.2 |
| |||
ecs.g6.13xlarge | 52 vCPUs + 192 GB memory | None | 3 |
|
3.6 |
| |||
2.4 |
| |||
ecs.g5.8xlarge | 32 vCPUs + 128 GB memory | None | 1.8 |
|
2.4 |
| |||
ecs.hfc6.16xlarge | 64 vCPUs + 128 GB memory | None | 3.6 |
|
2.4 |
| |||
ecs.g7.16xlarge | 64 vCPUs + 256 GB memory | None | 3.6 |
|
2.4 |
| |||
4.2 | China (Hong Kong) | |||
ecs.hfc6.20xlarge | 80 vCPUs + 192 GB memory | None | 4.8 |
|
4.2 |
| |||
3 |
| |||
ecs.g6.26xlarge | 104 vCPUs + 384 GB memory | None | 5.4 | India (Mumbai) |
6.6 |
| |||
4.2 |
| |||
7.2 | China (Hong Kong) | |||
ecs.g5.16xlarge | 64 vCPUs + 256 GB memory | None | 3 | India (Mumbai) |
4.2 |
| |||
3.6 |
| |||
4.8 |
| |||
ecs.gn6i-c4g1.xlarge | 4 vCPUs + 15 GB memory | 1 * NVIDIA T4 | 1.2 |
|
1.8 |
| |||
ecs.gn6i-c8g1.2xlarge | 8 vCPUs + 31 GB memory | 1 * NVIDIA T4 | 1.8 |
|
1.2 |
| |||
2.4 |
| |||
ecs.gn6i-c16g1.4xlarge | 16 vCPUs + 62 GB memory | 1 * NVIDIA T4 | 2.4 |
|
1.8 |
| |||
ecs.gn7-c12g1.3xlarge | 12 vCPUs + 95 GB memory | 1 * NVIDIA A100 | 3.6 | Malaysia (Kuala Lumpur) |
5.4 | China (Beijing) | |||
2.4 |
| |||
ecs.gn6i-c24g1.6xlarge | 24 vCPUs + 93 GB memory | 1 * NVIDIA T4 | 2.4 |
|
3 |
| |||
ecs.gn7i-c32g1.8xlarge | 32 vCPUs + 188 GB memory | 1 * NVIDIA A10 | 6.6 | India (Mumbai) |
3.6 |
| |||
3 | Indonesia (Jakarta) | |||
4.8 |
| |||
12.6 | China (Hong Kong) | |||
2.4 | China (Shenzhen) | |||
7.2 | Germany (Frankfurt) | |||
ecs.gn7e-c16g1.4xlarge | 16 vCPUs + 125 GB memory | 1 * NVIDIA A100 | 6.6 | Indonesia (Jakarta) |
4.8 | China (Beijing) | |||
9.6 |
| |||
2.4 | China (Hong Kong) | |||
ecs.gn7i-c16g1.4xlarge | 16 vCPUs + 60 GB memory | 1 * NVIDIA A10 | 3 |
|
7.2 | Malaysia (Kuala Lumpur) | |||
6 | China (Beijing) | |||
40.8 |
| |||
2.4 | China (Shenzhen) | |||
ecs.gn6i-c24g1.12xlarge | 48 vCPUs + 186 GB memory | 2 * NVIDIA T4 | 4.8 |
|
5.4 |
| |||
ecs.gn7i-c32g1.16xlarge | 64 vCPUs + 376 GB memory | 2 * NVIDIA A10 | 13.2 | India (Mumbai) |
6.6 | Singapore | |||
3 |
| |||
6 |
| |||
14.4 | Germany (Frankfurt) | |||
ecs.gn7-c13g1.13xlarge | 52 vCPUs + 380 GB memory | 4 * NVIDIA A10 | 3 |
|
12.6 | China (Beijing) | |||
ecs.gn7i-c32g1.32xlarge | 128 vCPUs + 752 GB memory | 4 * NVIDIA A10 | 3 |
|
13.8 | Singapore | |||
2.4 |
| |||
3.6 |
| |||
ecs.gn6i-c24g1.24xlarge | 96 vCPUs + 372 GB memory | 4 * NVIDIA T4 | 10.2 |
|
10.8 |
| |||
9.6 | China (Hong Kong) | |||
ecs.gn6v-c10g1.20xlarge | 82 vCPUs + 336 GB memory | 8 * NVIDIA V100 | 35.4 |
|
ecs.gn7-c13g1.26xlarge | 104 vCPUs + 760 GB memory | 8 * NVIDIA A100 | 3 |
|
3.6 |
| |||
ecs.ebmgn7e.32xlarge | 128 vCPUs + 1,024 GB memory | 8 * NVIDIA A100 | 45 |
|
50.4 | Germany (Frankfurt) |