Using Cambricon GPUs¶

This document describes how to use Cambricon GPUs in DCE 5.0.

Prerequisites¶

• DCE 5.0 container management platform has been deployed and is running properly.
• The container management module has either connected to a Kubernetes cluster or created a Kubernetes cluster, and the UI interface of the cluster is accessible.

• The current cluster has installed the Cambricon firmware, drivers, and DevicePlugin component. For installation details, please refer to the official documentation:

  • Driver and Firmware Installation
  • DevicePlugin Installation

Note: When installing the DevicePlugin, make sure to disable the --enable-device-type parameter; otherwise, DCE 5.0 will not be able to recognize Cambricon GPUs correctly.

Cambricon GPU Modes¶

Cambricon GPUs support the following modes:

• Full GPU mode: Registers the entire Cambricon GPU to the cluster for use as a whole.
• Share mode: Allows a single Cambricon GPU to be shared among multiple Pods. The number of shareable containers can be configured using the virtualization-num parameter.
• Dynamic SMLU mode: Provides fine-grained resource allocation, allowing control over the amount of GPU memory and compute assigned to each container.
• MIM mode: Splits a Cambricon GPU into multiple fixed-spec virtual GPUs for use.

Using Cambricon in DCE 5.0¶

Taking Dynamic SMLU mode as an example:

1. After correctly installing the DevicePlugin and related components, go to Cluster -> Cluster Maintenance -> Cluster Settings -> Addon Plugins to check whether the corresponding GPU type has been automatically enabled and detected.

   

2. Go to the Node Management page to verify whether the node has correctly recognized the GPU type.

   

3. Deploy workloads. Navigate to Cluster -> Workloads, and deploy workloads using an image. After selecting the type (e.g., MLU VGPU), configure the GPU resources required by the app:

   • GPU Compute Power (cambricon.com/mlu.smlu.vcore): Indicates the percentage of core compute the Pod needs.
   • GPU Memory (cambricon.com/mlu.smlu.vmemory): Indicates the amount of GPU memory the Pod needs, in MB.

   

YAML Configuration Example¶

Sample YAML file:

apiVersion: v1  
kind: Pod  
metadata:  
  name: pod1  
spec:  
  restartPolicy: OnFailure  
  containers:  
    - image: ubuntu:16.04  
      name: pod1-ctr  
      command: ["sleep"]  
      args: ["100000"]  
      resources:  
        limits:  
          cambricon.com/mlu: "1" # use this when device type is not enabled, else delete this line.  
          #cambricon.com/mlu: "1" #uncomment to use when device type is enabled  
          #cambricon.com/mlu.share: "1" #uncomment to use device with env-share mode  
          #cambricon.com/mlu.mim-2m.8gb: "1" #uncomment to use device with mim mode  
          #cambricon.com/mlu.smlu.vcore: "100" #uncomment to use device with smlu mode  
          #cambricon.com/mlu.smlu.vmemory: "1024" #uncomment to use device with smlu mode

Comments