Application Scenarios and Solutions
Applicable Hardware Models
Table 1 lists the Ascend hardware devices that support the AVI. For details about component features, see the corresponding sections.
Device Type |
Device Model |
|---|---|
Training device (configured with the Ascend 910) |
Atlas 900 PoD (model 9000) |
Atlas 900T PoD Lite |
|
Atlas 800 training server (model 9000) |
|
Atlas 800 training server (model 9010) |
|
Atlas 300T training card |
|
Atlas 300T Pro training card |
|
Inference device (configured with Atlas inference products) |
Atlas 300I Pro inference card |
Atlas 300I Duo inference card |
|
Atlas 300V video analysis cardA300V video analysis card |
|
Atlas 300V Pro video analysis card |
Application Scenarios
The AVI is suitable for scenarios with concurrent tasks running by multiple users, and each task has low computing power requirement.
Application Solutions
In the Ascend solution, the AVI now supports the following application scenarios:
- Native Docker: used together with the native Docker. You can use the npu-smi tool to create multiple vNPUs and use the Docker to mount the vNPUs to a container when it starts the container.
- Ascend Docker Runtime: used together with the Ascend Docker Runtime (container engine plugin). You can use the npu-smi tool to create multiple vNPUs and use Ascend Docker to mount the vNPUs a container when it starts and runs the container.
- Cluster scheduling component: works with cluster scheduling components Ascend Device Plugin and Volcano of MindX DL to support static virtualization.
In static virtualization mode, you can use the npu-smi tool to create multiple vNPUs in advance. When vNPUs are required, you can allocate vNPUs to upper-layer users based on the device discovery, allocation, and health status reporting functions of the Ascend Device Plugin component. In this solution, the Volcano component of the cluster scheduling component is optional.