Introduction
HAMi is the value-added accelerator path delivered by Alauda for ACP clusters. Use it when multiple workloads need to share accelerator devices, when workloads need fractional accelerator capacity instead of a whole physical device, or when a deployment needs HAMi scheduling and virtualization behavior on top of a vendor base path.
HAMi is a community project that supports multiple accelerator backends. The Alauda documentation focuses on the product packaging, ACP installation path, supported backends, resource keys, monitoring entry points, and troubleshooting steps that users need when running HAMi in ACP.
HAMi™ is a trademark of The Linux Foundation. Alauda is an independent vendor. This product is not affiliated with, endorsed by, or sponsored by The Linux Foundation. All trademarks are the property of their respective owners and are used here for identification purposes only.
When to use HAMi
Use Alauda AI Infrastructure Accelerator Virtualization when you need one or more of the following capabilities:
- share one accelerator device across multiple workloads;
- request GPU count, compute ratio, and memory independently on supported backends;
- use HAMi on NVIDIA GPU, HAMi on Ascend NPU, or HAMi on Ascend vNPU when those backends are supported by the target release;
- use ACP quota and application pages with HAMi resource names;
- monitor shared accelerator usage through ACP monitoring or HAMi WebUI;
- manage the HAMi scheduler, backend-specific device exposure, and runtime integration as ACP-delivered components.
For NVIDIA GPU allocation without HAMi sharing semantics, use the NVIDIA GPU vendor base documentation. For direct Ascend NPU allocation without HAMi, continue with Request Ascend NPU resources directly.
For combined deployments, use the vendor base documentation to prepare nodes, drivers, runtimes, and CDI. Then use the HAMi documentation to install the components required by the selected HAMi scenario, request HAMi resources, and troubleshoot scheduling or sharing.
How backend components fit together
Alauda AI Infrastructure Accelerator Virtualization provides the HAMi scheduler, webhook, runtime integration, and a built-in NVIDIA device plugin through an OLM-managed Operator. Ascend device exposure is provided by a separately installed product when the selected version combination is supported.
HAMi-WebUI is a separately installed optional component that provides a HAMi-specific resource overview.
Product scope
This documentation covers the Alauda delivery of HAMi in ACP:
- how to install and verify the HAMi Operator;
- how to install and configure HAMi-WebUI and the backend-specific HAMi Ascend Device Plugin;
- how to choose HAMi on NVIDIA GPU, HAMi on Ascend NPU, or HAMi on Ascend vNPU;
- how to request HAMi resources in workloads;
- how HAMi backend paths depend on NVIDIA GPU or Ascend NPU base prerequisites;
- how HAMi resource names connect to ACP quota and monitoring pages;
- which components, optional plugins, and versions are delivered by the product;
- common operational checks and troubleshooting steps.
For full upstream HAMi design details or backend-specific device behavior, continue with the HAMi community documentation or the corresponding hardware vendor documentation.
Main Tasks
- Choose a HAMi deployment scenario
- Install
Alauda AI Infrastructure Accelerator Virtualization - Install
Alauda Build of HAMi Ascend Device Plugin - Install
Alauda AI Infrastructure Accelerator Virtualization Console - Verify the installation
- Request HAMi resources
- Configure dynamic MIG
- Monitor HAMi usage
- Upgrade or uninstall HAMi components
- Check version and component support
- Read release notes