vNPU Usage Description

In Kubernetes scenarios, when you need to use vNPUs, you can use Ascend Device Plugin to enable Kubernetes to manage Ascend AI processor resources. You can use static virtualization and dynamic virtualization based on whether vNPUs need to be created in advance. These two virtualization modes cannot be used together nor used with the using method on Ascend Docker Runtime mentioned previously. The following table lists the product models supported by the Ascend virtual instance feature.

For details about how to install Ascend Device Plugin, see Ascend Device Plugin.

For details about optional components in the static virtualization scenario, see the following:

• Volcano: If you use a custom scheduling component, you need to set parameters. For details, see Table 2. Alternatively, you can use this component for scheduling.
• Ascend Operator: This component is required only when Atlas training product are used. The Atlas inference product do not require this component.