NPU

Use this page when you want to run container workloads on Ascend NPUs.

Choose an NPU path

Use the Ascend vendor integration for hardware enablement, driver lifecycle, runtime integration, CDI, direct NPU allocation, and vendor-standard operations. Use HAMi when the user task starts from sharing, virtualization, or scheduling behavior built on an Ascend base.

If a deployment prepares Ascend nodes with NPU Operator but uses HAMi Ascend Device Plugin to expose resources, use the NPU product documentation for base prerequisites and the HAMi product documentation for device exposure, resource requests, sharing semantics, and troubleshooting.

Use NPU resources

  • Workloads request NPU resources exposed by the selected allocation product.
  • Use project and namespace pages for quota operations.
  • If an NPU resource is allocatable on the node but not visible in ACP quota pages, register the accelerator quota field metadata in Accelerator resource quota.
  • Use observability pages for dashboard and alert configuration.
  • Use NPU product documentation for Ascend base resource names, driver lifecycle, and validation workloads. Use HAMi documentation for HAMi-specific Ascend resource names or sharing behavior.

Product documentation

Use the NPU product site for installation, driver lifecycle, compatibility, release notes, and troubleshooting.

Note
Because Alauda Ascend NPU releases on a different cadence from Alauda Container Platform, the Alauda Ascend NPU documentation is now available as a separate documentation set at Alauda Ascend NPU.