The purpose of this document is to provide a configuration guide for demonstrating computer vision inferencing on an Ampere CPU using the HPE ProLiant RL300 Gen11 Compute server, running Ubuntu and Canonical MicroK8s, a lightweight Kubernetes distribution.

The combination of Ampere® Altra® family processors and Canonical MicroK8s is ideal for AI inference workloads, offering a scalable platform from single node setups to high-availability multi-node production clusters, while minimizing the physical footprint.

Figure 1. The framework for Computer vision AI Inferencing demo with HPE ProLiant RL300 Gen11

Software versions

• Ubuntu Server: 24.04
• MicroK8s: 1.31
• YOLOv8 Demo Application: 0.3.9

Ubuntu Server

Ubuntu Server is a version of the Ubuntu operating system specifically designed for server and data center environments, offering a robust and scalable platform for cloud, edge, and on-premises infrastructure. It supports a wide range of workloads, from web hosting and database management to AI, machine learning, and containerized applications. Ubuntu Pro is a subscription-based offering that extends the standard Ubuntu distribution with additional enterprise-grade features, including security and compliance enhancements. Organizations using Ubuntu Pro benefit from extended security maintenance coverage for up to 10 years, ensuring long-term stability and protection against vulnerabilities. In addition, Ubuntu Pro provides optional enterprise-grade phone and ticket support from Canonical, offering companies the expertise and assistance they need to manage and optimize their infrastructure. This makes Ubuntu Pro a comprehensive solution for businesses seeking enhanced security, reliability, and professional support in their Ubuntu deployments.

MicroK8s

Canonical MicroK8s is a CNCF-certified Kubernetes distribution that provides a lightweight, efficient, and streamlined approach to deploying and managing Kubernetes clusters. Designed with simplicity and speed in mind, it caters to developers, IoT, and edge computing environments, as well as production workloads in resource-constrained environments. Delivered as a snap, the universal Linux app packaging format, MicroK8s dramatically simplifies installation, updates, and upgrades of its components, ensuring a hassle-free deployment process with minimal overhead.

MicroK8s offers a fully-featured Kubernetes experience with key services like DNS, storage, ingress, and monitoring enabled out of the box, while allowing users to enable or disable additional services through simple commands. Its modular design allows for flexibility in tailoring the cluster to specific needs without the complexity of a full-scale Kubernetes deployment. MicroK8s is also well-suited for AI, ML, and edge workloads, making it a versatile choice for a wide range of use cases, from development and testing to production environments.

YOLOv8

YOLO is a real-time object detection algorithm designed to identify and classify objects within an image or video in a single pass through the neural network, making it highly efficient and fast. It works by dividing the image into a grid and predicting bounding boxes and class probabilities for each section. YOLOv8, the latest iteration, enhances detection accuracy and speed, making it ideal for AI-powered tasks such as autonomous driving and surveillance. When deployed on MicroK8s, a lightweight Kubernetes distribution, YOLO can efficiently scale from edge devices to production clusters, leveraging MicroK8s' container orchestration and AI inferencing capabilities to run YOLO's deep learning models in a cloud-native environment. This ensures optimized performance and easy deployment of AI workloads across various infrastructures.

Running YOLO on an Ampere Altra CPU with the HPE RL300 server brings significant performance gains, as the high core counts and energy efficiency of the Ampere CPU optimize AI inferencing tasks. This combination provides a scalable and cloud-native platform for running advanced AI workloads while minimizing physical and energy footprints.

Accessing the Demo

To access the YOLOv8 object detection inference demo, observability and Kubernetes dashboards, following FQDNs need to be resolvable to the IP address of the RL300 by the machine running the browser. Replace with the IP address of the RL300.

In Linux, these entries need to be added to the /etc/hosts file.

In Windows, these entries need to be added to the c:\Windows\System32\Drivers\etc\hosts file. Right click on Notepad in your start menu and select “Run as Administrator”. Then, click File > Open and browse to : c:\Windows\System32\Drivers\etc\hosts and add the entries.

Alternatively, you can add these entries to the DNS server that the machine running the browser uses.

kubernetes.microk8s.test grafana.microk8s.test prometheus.microk8s.test alertmanager.microk8s.test yolo.microk8s.test

Install and update Ubuntu 24.04

• Download and install Ubuntu 24.04
• Update system

Copy
sudo apt update && sudo apt upgrade -y

Install MicroK8s

Install MicroK8s package

Copy
sudo snap install microk8s --channel=1.31-strict/stable

Add your user to the ‘microk8s’ group for unprivileged access

Copy
sudo adduser $USER snap_microk8s

Wait for MicroK8s to finish initializing

Copy
sudo microk8s status --wait-ready

Enable the 'storage' and 'dns' addons

Copy
sudo microk8s enable hostpath-storage dns

Alias kubectl so it interacts with MicroK8s by default

Copy
sudo snap alias microk8s.kubectl kubectl

Ensure your new group membership is apparent in the current terminal (Not required once you have logged out and back in again)

Copy
newgrp snap_microk8s

Apply yolov8-deployment.yaml file to deploy YOLOv8 inference demo

Copy
cat << EOF > yolov8-deployment.yaml
kind: Service
apiVersion: v1
metadata:
  name: yolov8-app1-svc
  labels:
    app: yolov8-app1
spec:
  ports:
    - protocol: TCP
      port: 80
      targetPort: 5000
  selector:
    app: yolov8-app1
  type: ClusterIP
  sessionAffinity: None


---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: yolov8-app1
spec:
  selector:
    matchLabels:
      app: yolov8-app1 
  replicas: 1
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        # Label is used as selector in the service.
        app: yolov8-app1
    spec:
      nodeSelector:
        node-role: node1
      containers:
      - name: yolov8-app1
        image: ghcr.io/amperecomputingai/ampere-ai-ref-apps:yolov8-0.3.9
        env:
        - name: SUBTITLE_OVERRIDE
          value: "1"
        - name: GRADIO_SERVER_NAME
          value: "0.0.0.0"
        - name: GRADIO_SERVER_PORT
          value: "5000"
        - name: NTHREADS 
          value: "16"
        - name: NSTREAMS
          value: "2"
        - name: VIDEO_SRC
          value: ""
        - name: WEBCAM0_SRC 
          value: "0"
        - name: WEBCAM1_SRC 
          value: "2"
        - name: CONFIG_FILE
          value: 'cfg/config.yaml'
        ports:
        - containerPort: 5000 
          name: yolov8-app1
        resources:
          limits:
            cpu: "192"
          requests:
            cpu: "32"
        volumeMounts:
        - mountPath: /dev/video0
          name: dev-video0
        - mountPath: /dev/video2
          name: dev-video2
        securityContext:
          privileged: true
      volumes:
      - name: dev-video0
        hostPath:
          path: /dev/video0
      - name: dev-video2
        hostPath:
          path: /dev/video2
            #nodeSelector:
            #accept-ai-pod: yolov8-app


---
kind: Ingress
apiVersion: networking.k8s.io/v1
metadata:
  name: yolov8-app-ingress
  annotations:
    kubernetes.io/ingress.class: public
spec:
  rules:
  - host: "yolo.microk8s.test" 
    http:
      paths:
      - path: /
        pathType: Prefix
        backend:
          service:
            name: yolov8-app1-svc
            port:
              number: 80
EOF

Copy
kubectl apply -f yolov8-deployment.yaml

This demo will showcase 2 inference streams running simultaneously.

Access dashboard web UI via https://yolo.microk8s.test

1. Select the source from ‘Stream - Video Source’ Dropdown menu 2. Click ‘Start’ button to start the demo