Lingjun is a high-performance computing resource offered by PAI for large-scale AI training and inference workloads. It supports the DSW, DLC, and EAS modules. Developed by Alibaba Cloud, Lingjun provides RDMA high-speed networking, high-performance communication libraries, acceleration software, and GPU virtualization to meet demanding compute requirements. PAI offers fully managed Lingjun resources that you can purchase and use directly from the console. If you have already purchased Lingjun hardware separately, you can also add it as a semi-managed resource. You must create a resource group before purchasing resources.
Limits
-
Regions
Lingjun is available only in the China (Ulanqab), Singapore, China (Shenzhen), China (Beijing), China (Shanghai), and China (Hangzhou) regions.
-
Eligibility
Lingjun supports both subscription and pay-as-you-go billing methods. This feature is available only to allowlisted users. Contact your account manager to apply for access.
-
Job types
Only the following DLC training job types are supported: TensorFlow, PyTorch, ElasticBatch, XGBoost, OneFlow, MPIJob, Slurm, and Ray.
-
Account and permission requirements
Alibaba Cloud account (recommended): No additional authorization is required. You can perform all operations directly.
-
RAM user: The Alibaba Cloud account must grant the resource pool management permissions or the AliyunPAIFullAccess permission to the RAM user.
ImportantAliyunPAIFullAccess grants full permissions to manage all PAI resources and features. Exercise caution when granting this policy. We recommend that you use the Alibaba Cloud account to perform operations.
Create a resource group
Log on to the PAI console and go to the AI Computing Resources > Resource Pools page.
On the Lingjun Intelligent Computing resources tab, click Create Resource Group.
-
In the Create Resource Group dialog box, configure the parameters and click OK.
Parameter
Description
Type
Select Dedicated Resource Group.
Resource Group Name
Enter a name for the resource group as prompted.
High-bandwidth interconnect zone
Lingjun organizes underlying compute resources into high-bandwidth interconnect zones (hz). Resources within the same hz can communicate over high-bandwidth networking. When purchasing, allocating, or using resources, you can customize the hz affinity. After you enable this option, the resource group can only purchase compute resources from the selected hz.
Purchase Lingjun resources
Lingjun resources support both subscription and pay-as-you-go billing methods. After you create a resource group, you can purchase resources. For billing details, see Billing of AI computing resources.
Purchase subscription resources
In the resource group list on the Intelligent Computing LINGJUN Resources tab, click Create Order > Create Subscription Order in the Actions column. Alternatively, click the resource group name and then click Create Order > Create Subscription Order in the upper-right corner of the resource group details page.
-
On the buy page, the resource group ID and region are preselected. Select the Node Specification, Quantity, Subscription Duration, and other parameters, and then click Buy Now.
Common issues and solutions:
-
The order does not contain information for the current resource group.
Cause: After switching regions, the resource group ID and region do not match.
Solution: Switch back to the default region of the resource group.
-
The specified instance type is out of stock in zone.
Cause: The selected node type is out of stock in the specified region.
Solution: Select a different node type.
-
The current kind of instance is temporarily not supported. Please choose another kind of ecs to purchase.
Cause: The selected node type is not supported (unavailable for purchase) in the specified region.
Solution: Select a different node type.
-
-
After payment is complete, the purchased resources appear on the order list page. On the resource group details page, click the Order Information tab to view order instance details, including resource specifications, quantity, running status, GPU type, CPU cores, memory, creation time, and expiration time.
Click Renew to renew a subscription. Orders are split by node instance for easier per-node management.
Purchase pay-as-you-go resources
Pay-as-you-go billing is available only to allowlisted users. To apply for access, contact your account manager.
In the resource group list on the Lingjun Intelligent Computing resources tab, click Create Order > Create Pay-As-You-Go Order in the Actions column. Alternatively, click the resource group name and then click Create Order > Create Pay-As-You-Go Order in the upper-right corner of the resource group details page.
-
On the Create Pay-As-You-Go Order page, the Region, Resource Type, and Resource Group are preselected.
Select the Resource Specification and configure the node Quantity, and then click OK.
-
After the purchase is complete, you can filter by billing method to distinguish between subscription and pay-as-you-go orders.
ImportantWhen you no longer need a pay-as-you-go resource, click Release in the Actions column on the Order Information page to stop incurring charges. After a node is released, all jobs running on that node are terminated. Proceed with caution.
What to do next
After you create a resource group and purchase compute resources, you can:
View basic information and manage purchased resources on the resource group page. For more information, see Manage resources.
Configure resource quotas and allocate resources to training jobs. For more information, see Create a resource quota.