文档

Setting the NPU Sharing Mode

The NPU sharing mode is supported by the Atlas 500 AI edge station (model 3000) and Atlas 200 AI accelerator module (RC). That is, multiple containers can share the NPU resources of the same processor. In this way, a processor can house multiple containers to run different inference services. By default, the NPU sharing mode is disabled. You can enable this mode by performing the following operations.

  • The NPU sharing mode of the Atlas 500 AI edge station (model 3000) supports configuration persistency.
  • The NPU sharing mode of the Atlas 200 AI accelerator module (RC) does not support configuration persistence. After the device is reset, the mode needs to be reconfigured.
  • After the NPU sharing mode is disabled, a container occupies the NPU exclusively, which affects the running of the container. Exercise caution when performing this operation.
  • If the version is rolled back to MindX Edge V100R020C10 or an earlier version that does not support the NPU sharing mode, multiple containers that have been deployed before the rollback and share the same NPU resource may not run properly after the rollback. Exercise caution when performing this operation.
Using FusionDirector:
  1. On the FusionDirector WebUI, choose Menu > Devices > Device List > Edge Devices to go to the page for managing edge devices.
  2. In the edge device list, select one or more devices for which you want to enable the NPU sharing mode, choose More > NPU Sharing Mode in the upper right corner of the page, and click Enable.

    The task for enabling the mode is displayed in the task list in the upper right corner of the page.

    To disable the NPU sharing mode, click Disable. The disabling task is displayed in the task list in the upper right corner of the page.

Parent topic: Common Operations