Use compute acceleration on Azure Stack Edge Pro GPU for Kubernetes deployment

APPLIES TO: Azure Stack Edge Pro - GPUAzure Stack Edge Pro 2Azure Stack Edge Pro RAzure Stack Edge Mini R                  

This article describes how to use compute acceleration on Azure Stack Edge devices when using Kubernetes deployments.

About compute acceleration

The Central Processing Unit (CPU) is your default general purpose compute for most processes running on a computer. Often a specialized computer hardware is used to perform some functions more efficiently than running those in the software in a CPU. For example, a Graphics Processing Unit (GPU) can be used to accelerate the processing of pixel data.

Compute acceleration is a term used specifically for Azure Stack Edge devices where a Graphical Processing Unit (GPU), a Vision Processing Unit (VPU), or a Field Programmable Gate Array (FPGA) are used for hardware acceleration. Most workloads deployed on your Azure Stack Edge device involve critical timing, multiple camera streams, and/or high frame rates, all of which require specific hardware acceleration.

The article will discuss compute acceleration only using GPU or VPU for the following devices:

• Azure Stack Edge Pro GPU - These devices can have 1 or 2 Nvidia T4 Tensor Core GPU. For more information, see NVIDIA T4.
• Azure Stack Edge Pro R - These devices have 1 Nvidia T4 Tensor Core GPU. For more information, see NVIDIA T4.
• Azure Stack Edge Mini R - These devices have 1 Intel Movidius Myriad X VPU. For more information, see Intel Movidius Myriad X VPU.

Use GPU for Kubernetes deployment

The following example yaml can be used for an Azure Stack Edge Pro GPU or an Azure Stack Edge Pro R device with a GPU.

apiVersion: v1
kind: Pod
metadata:
  name: gpu-pod
spec:
  containers:
    - name: cuda-container
      image: nvidia/cuda:9.0-devel
      resources:
        limits:
          nvidia.com/gpu: 1 # requesting 1 GPU
    - name: digits-container
      image: nvidia/digits:6.0
      resources:
        limits:
          nvidia.com/gpu: 1 # requesting 1 GPU

Use VPU for Kubernetes deployment

The following example yaml can be used for an Azure Stack Edge Mini R device that has a VPU.

apiVersion: batch/v1
kind: Job
metadata:
 name: intelvpu-demo-job
 labels:
   jobgroup: intelvpu-demo
spec:
 template:
   metadata:
     labels:
       jobgroup: intelvpu-demo
   spec:
     restartPolicy: Never
     containers:
       -
         name: intelvpu-demo-job-1
         image: ubuntu-demo-openvino:devel
         imagePullPolicy: IfNotPresent
         command: [ "/do_classification.sh" ]
         resources:
           limits:
             vpu.intel.com/hddl: 1

Next steps

Learn how to Use kubectl to run a Kubernetes stateful application with a PersistentVolume on your Azure Stack Edge Pro GPU device.