Updated on 2024-08-16 GMT+08:00

xGPU View

A GPU can be virtualized into xGPUs. The xGPU view displays the GPU memory and compute allocation rate of xGPUs from multiple perspectives, such as nodes, GPUs, and containers, so you can better monitor the GPU.

Metric Description

Figure 1 Resource metrics
Table 1 Resource metrics

Metric

Unit

Description

Cluster - xGPU Device GPU Memory Usage

%

GPU memory usage of all xGPU devices in the cluster.

Formula: Used GPU memory of all xGPU devices in the cluster/Total GPU memory of the cluster

Cluster - xGPU Device GPU Compute Usage

%

GPU compute usage of all xGPU devices in the cluster.

Formula: Used GPU compute of all xGPU devices in the cluster/Total GPU compute of the cluster

Node - xGPU Device GPU Memory Usage

%

GPU memory usage of all xGPU devices on the node.

Formula: Used GPU memory of all xGPU devices on the node/Total GPU memory of the node

Node - xGPU Device GPU Compute Usage

%

GPU compute usage of all xGPU devices on the node.

Formula: Used GPU compute of all xGPU devices on the node/Total GPU compute of the node

Node - Number of xGPU Devices

N/A

The number of xGPU devices on the node.

Node - Allocated GPU Memory of xGPU Devices

byte

Maximum GPU memory that can be used by all xGPU devices on the node.

GPU - xGPU Device GPU Memory Usage

%

Percentage of used GPU memory of all xGPU devices on the GPU.

Formula: Used GPU memory of all xGPU devices on the GPU/Total GPU memory of the GPU

GPU - Allocated GPU Memory of xGPU Devices

byte

Maximum GPU memory that can be used by all xGPU devices on the GPU.

GPU - GPU Memory Allocation Rate of xGPU Devices

%

Percentage of the maximum GPU memory that can be used by all xGPU devices on the GPU.

Formula: Maximum GPU memory that can be used by all xGPU devices on the GPU/Total GPU memory of the GPU

GPU - xGPU Device GPU Compute Usage

%

Percentage of the GPU compute used by all xGPU devices on the GPU.

Formula: GPU compute used by all xGPU devices on the GPU/Total GPU compute of the GPU

GPU - Number of xGPU Devices

N/A

The number of xGPU devices created on the GPU.

GPU - Scheduling Policy

N/A

  • 0: xGPU memory is isolated and cores are shared.
  • 1: Both xGPU memory and cores are isolated.
  • 2: default mode, indicating that the current card is not used by any xGPU device for allocation.

GPU - Number of Unhealthy xGPU Devices

N/A

The number of unhealthy xGPU devices on the GPU.

Allocated Container GPU Memory

byte

Maximum GPU memory that can be used by a container.

Container GPU Compute Usage

%

GPU compute used by containers on the xGPU device.

Formula: GPU compute used by containers on the xGPU device/Total GPU compute of the xGPU device

Used Container GPU Memory

byte

GPU memory used by containers on the xGPU device.

Container GPU Memory Usage

%

GPU compute used by containers on the xGPU device.

Formula: GPU memory used by containers on the xGPU device/Total GPU memory of the xGPU device

Metric List

The following is the metric list of the xGPU view.
Table 2 Metric description

Metric

Type

Description

xgpu_memory_total

Gauge

Total xGPU memory.

xgpu_memory_used

Gauge

Used xGPU memory.

xgpu_core_percentage_total

Gauge

Total xGPU compute.

xgpu_core_percentage_used

Gauge

xGPU compute usage.

gpu_schedule_policy

Gauge

There are three GPU modes.

  • 0: xGPU memory is isolated and cores are shared.
  • 1: Both xGPU memory and cores are isolated.
  • 2: default mode, indicating that the current card is not used by any xGPU device for allocation.

xgpu_device_health

Gauge

Health of an xGPU device. Currently, the virtualization domain does not provide a specific interface to check the xGPU health. The xGPU health is based on the health of the GPU where the xGPU device is located. The value 0 indicates that the xGPU is healthy, and the value 1 indicates that the xGPU is unhealthy.