文档

Computing Power Allocation of the Atlas 300I Pro Inference Card Supported in Edge Scenarios

For details about the existing modes of computing power allocation, see "Virtualization Rules > Virtualization Modes" in the Ascend Virtualization Instance (AVI) User Guide.

When NPU resources are divided into vNPUs or some of multiple NPU resources are divided into vNPUs, the AtlasEdge identifies resources of only a certain specification.

Identification Rule

  • When there are multiple types of resources, the AtlasEdge identifies and reports the resource with the largest quantity.
  • When the number of two or more types of resources is the same, the AtlasEdge identifies and reports the resource with the largest capability.

Examples

Example 1: If there is only one NPU (one Atlas 300I Pro inference card) and it is divided into vir02, vir02, vir02, and vir02_1c, three vir02 resources are identified and reported.

Example 2: If there are five NPUs (five Atlas 300I Pro inference cards) and one NPU is divided into vir02, vir02, vir02, and vir02_1c, four NPUs are identified and reported due to the NPU quantity.