NPU Metrics
If the CCE AI Suite (Ascend NPU) add-on version is 2.1.55 or later, NPU-Exporter can be used to monitor and collect Ascend AI processor metrics. NPU-Exporter is designed to obtain and report runtime data of Ascend AI chips, including the number of Ascend AI processors and real-time network port receiving rate. These metrics are referred to as NPU metrics. By monitoring NPU metrics, you can gain real-time visibility into NPU performance, detect and resolve potential issues, and ensure that NPUs run stably and efficiently. This section provides a detailed overview of the NPU metrics reported by NPU-Exporter.
Applicable NPU Nodes
Only NPU metrics of the nodes listed in AI-accelerated ECSs can be monitored and collected.
NPU Metrics
NPU-Exporter can collect 73 NPU metrics. This section focuses on the common metrics supported by nodes listed in AI-accelerated ECSs. For details, see Table 1.
| Category | Metric | Description | Metric Label | Field Type |
|---|---|---|---|---|
| NPU | npu_chip_info_name | Name and ID of an Ascend AI processor | container_name: a container name | String |
| id: an NPU ID | String | |||
| model_name: name of an Ascend AI processor | String | |||
| namespace: a namespace name | String | |||
| pcie_bus_info: PCIe information of an Ascend AI processor | String | |||
| pod_name: a pod name | String | |||
| vdie_id: Unique ID of an Ascend AI processor, which can be used as the UUID of the NPU | String | |||
| npu_chip_info_health_status | Health status of an Ascend AI processor. Options:
| container_name: a container name | String | |
| id: an NPU ID | String | |||
| model_name: name of an Ascend AI processor | String | |||
| namespace: a namespace name | String | |||
| pcie_bus_info: PCIe information of an Ascend AI processor | String | |||
| pod_name: a pod name | String | |||
| vdie_id: Unique ID of an Ascend AI processor, which can be used as the UUID of the NPU | String | |||
| npu_chip_info_power | Power consumption of an Ascend AI processor, in watts (W) NOTE: If the NPU on the node is Snt3P, this metric specifies the board power consumption. If the NPU is Snt3, this metric specifies the power consumption of the Ascend AI processor. | container_name: a container name | String | |
| id: an NPU ID | String | |||
| model_name: name of an Ascend AI processor | String | |||
| namespace: a namespace name | String | |||
| pcie_bus_info: PCIe information of an Ascend AI processor | String | |||
| pod_name: a pod name | String | |||
| vdie_id: Unique ID of an Ascend AI processor, which can be used as the UUID of the NPU | String | |||
| npu_chip_info_temperature | Temperature of an Ascend AI processor, in degrees Celsius (°C) | container_name: a container name | String | |
| id: an NPU ID | String | |||
| model_name: name of an Ascend AI processor | String | |||
| namespace: a namespace name | String | |||
| pcie_bus_info: PCIe information of an Ascend AI processor | String | |||
| pod_name: a pod name | String | |||
| vdie_id: Unique ID of an Ascend AI processor, which can be used as the UUID of the NPU | String | |||
| npu_chip_info_utilization | AI Core usage of an Ascend AI processor, in percentage | container_name: a container name | String | |
| id: an NPU ID | String | |||
| model_name: name of an Ascend AI processor | String | |||
| namespace: a namespace name | String | |||
| pcie_bus_info: PCIe information of an Ascend AI processor | String | |||
| pod_name: a pod name | String | |||
| vdie_id: Unique ID of an Ascend AI processor, which can be used as the UUID of the NPU | String | |||
| npu_chip_info_vector_utilization | AI Vector usage of an Ascend AI processor | container_name: a container name | String | |
| id: an NPU ID | String | |||
| model_name: name of an Ascend AI processor | String | |||
| namespace: a namespace name | String | |||
| pcie_bus_info: PCIe information of an Ascend AI processor | String | |||
| pod_name: a pod name | String | |||
| vdie_id: Unique ID of an Ascend AI processor, which can be used as the UUID of the NPU | String | |||
| DDR | npu_chip_info_used_memory | Used DDR memory of an Ascend AI processor, in MB | container_name: a container name | String |
| id: an NPU ID | String | |||
| model_name: name of an Ascend AI processor | String | |||
| namespace: a namespace name | String | |||
| pcie_bus_info: PCIe information of an Ascend AI processor | String | |||
| pod_name: a pod name | String | |||
| vdie_id: Unique ID of an Ascend AI processor, which can be used as the UUID of the NPU | String | |||
| npu_chip_info_total_memory | Total DDR memory of an Ascend AI processor, in MB | container_name: a container name | String | |
| id: an NPU ID | String | |||
| model_name: name of an Ascend AI processor | String | |||
| namespace: a namespace name | String | |||
| pcie_bus_info: PCIe information of an Ascend AI processor | String | |||
| pod_name: a pod name | String | |||
| vdie_id: Unique ID of an Ascend AI processor, which can be used as the UUID of the NPU | String |
Helpful Links
You can use NPU-Exporter to monitor these NPU metrics. For details, see Comprehensive Monitoring of NPU Metrics.