Comprehensive Upgrades to ModelArts Resource Pool Management Functions

Features of New-Version Dedicated Resource Pools

The new-version dedicated resource pool management is a comprehensive technology and product improvement. The main improvements are as follows:

• Single dedicated resource pool type for diverse purposes: Dedicated resource pools are no longer classified into those for development/training and those for service deployment. You can run both training and inference workloads in a dedicated resource pool. You can also set the job types supported by a dedicated resource pool based on your needs.
• Dedicated resource pool network connection: You can create and manage dedicated resource pool networks on the ModelArts management console. If you need to access resources in your VPC for jobs running in a dedicated resource pool, interconnect the VPC with the dedicated resource pool network.
• More cluster details: The new-version dedicated resource pool details page provides more cluster details, such as jobs, nodes, and resource monitoring, helping you learn about the cluster status and better plan and use resources.
• Cluster GPU/NPU driver management: On the new-version dedicated resource pool details page, you can select an accelerator card driver and perform change upon submission or smooth upgrade of the driver based on service requirements.
• Fine-grained resource allocation (coming soon): You can divide your dedicated resource pool into multiple small pools and assign different quotas and permissions to each small pool for flexible and refined resource allocation and management.

More features will be provided in later versions for a better user experience.

Can I Continue to Use the Existing Dedicated Resource Pools After the Upgrade Takes Effect?

If you have created dedicated resource pools, you can still access the old-version dedicated resource pool (elastic cluster) management page on the ModelArts management console and use the created resource pools, but you cannot create dedicated resource pool on that page. ModelArts allows you to migrate existing dedicated resource pools to the new management page. You will be contacted to complete the migration and this does not require you to perform any operations. In addition, the migration does not affect the workloads running in the dedicated resource pools. Pay attention to the easy-to-use new management functions of dedicated resource pools. There is no change in creating training jobs or inference services.

Will New-Version Dedicated Resource Pools Be More Expensive?

The charging unit and unit price of the new-version dedicated resource pools are the same as those of the old-version dedicated resource pools. If you do not scale in or out your dedicated resource pools, the fee will not change. In addition, more value-added features, such as subpool division, elastic sharing, and data acceleration, will be provided in later versions to better allocate compute resources and improve cost-effectiveness.

Differences Between New and Old Dedicated Resource Pools

• In the old version, the dedicated resource pools dedicated for development/training are separated from those dedicated for service deployment. In addition, the pools of the two types offer different functions and their user experience varies. In the new version, the dedicated resource pools of the two types are unified. You only need to configure one or multiple job types. Then, the dedicated resource pool automatically supports the configured job type.
• New dedicated resource pools inherit all functions of the old ones and have greatly improved user experience in key functions such as purchasing and resizing a resource pool. Use new dedicated resource pools for smooth, transparent experience.
• Additionally, the new dedicated resource pools offer enhanced functions, for example, allowing you to upgrade GPU or Ascend drivers, view details about job queuing, and use one network for multiple pools. More new functions of the new dedicated resource pools are coming soon.

How Can I Get Help or Provide Feedback if I Encounter Problems During Use?

Similar to other ModelArts functions, you can report problems or obtain help in the sidebar of the console. In addition, you are advised to read the subsequent sections of this document to further understand how to use ModelArts dedicated resource pools. Submit a service ticket for more requirements.

Instructions of Dedicated Resource Pools

• If you use dedicated resource pools for the first time, get started by reading Resource Pool.
• Create a dedicated resource pool by referring to Creating a Resource Pool.
• View the details about a created dedicated resource pool by referring to Viewing Details About a Resource Pool.
• If the specifications of a dedicated resource pool do not meet your service requirements, adjust the specifications by referring to Resizing a Resource Pool.
• Set or change job types supported by a dedicated resource pool by referring to Changing Job Types Supported by a Resource Pool.
• Upgrade the GPU/Ascend driver of your dedicated resource pools by referring to Upgrading a Resource Pool Driver.
• If a dedicated resource pool is no longer needed, delete it by referring to Deleting a Resource Pool.
• If any exception occurs when you use a dedicated resource pool, handle the exception by referring to Abnormal Status of a Dedicated Resource Pool.
• Manage dedicated resource pool networks or interconnect VPCs with the networks by referring to ModelArts Network.