Introduction
Alauda Ascend NPU documentation covers the Ascend vendor base path for ACP accelerator workloads. It helps users prepare and operate Ascend NPU nodes, direct NPU allocation, RuntimeClass integration, monitoring components, and driver lifecycle workflows.
Alauda Build of NPU Operator is the core product in this base path. It installs and manages the Ascend NPU software stack on ACP workload clusters. It reconciles the Ascend driver, container runtime integration, device plugin, NPU exporter, and optional MindCluster components from an NPUOperatorCtl custom resource.
Use this product when ACP workloads need to request Ascend NPU resources such as huawei.com/Ascend910 for 910-family devices or huawei.com/Ascend310P, depending on what the installed device plugin reports in the target cluster.
This site currently covers the NPU Operator, Ascend Device Plugin, and the ascend RuntimeClass path backed by ascend-docker-runtime.
When to use it
Use Alauda Build of NPU Operator when:
- you run container workloads on Ascend 910B or Ascend 310P nodes;
- you want the Ascend driver, device plugin, runtime integration, and exporter to be managed as an ACP product;
- you need the operator to allocate NPU devices and configure the Ascend container runtime for workloads;
- you need a documented driver upgrade and node reboot workflow for managed NPU nodes.
Use HAMi documentation when the user task starts from sharing, virtualization, or scheduling behavior on Ascend devices, such as HAMi on Ascend NPU or HAMi on Ascend vNPU. In that combined path, this site provides Ascend base prerequisites; HAMi documentation owns HAMi installation, device exposure, resource keys, sharing semantics, and troubleshooting.
Use the ACP accelerator pages when you only need to choose between GPU, NPU, HAMi, DRA, and VM GPU passthrough paths. Use Huawei or openFuyao documentation for vendor hardware concepts, driver internals, and upstream API details that do not change ACP operation.