How to Deploy AMD Enterprise AI Platform on Vultr

AMD Enterprise AI Suite is a complete platform for building, deploying, and running AI workloads on Kubernetes tuned for AMD hardware. It can be used by system administrators, platform teams, AI researchers, and developers working on AI solutions.

This guide explains all the core components such as AMD AI Workbench, AMD Resource Manager, Kubernetes AI Workload Orchestrator (Kaiwo), Kubernetes Platform, Cluster Forge, and AMD Inference Microservices (AIMs) that are offered for AI compute use, you will also be able to deploy the AI platform using Vultr Cloud GPU Infrastructure and AMD Instinct™ MI300X GPUs.

Prerequisites

Before you begin, ensure you:

• Have access to an AMD Instinct™ MI300X GPU.
• Have a 2TB Block Storage volume attached for workloads.
• Have created a wildcard DNS A record pointing to your server's public IP address.

Key Components of the Platform

• AMD AI Workbench: Focuses on simplifying the execution of fine-tuning, inference or other jobs, enabling researchers to manage AI workloads by offering low-code approaches for developing AI applications. With a comprehensive model catalog and integrations with MLops tools such as MLflow, TensorBoard and Kubeflow, AMD AI Workbench allows researchers to use AI development tools in a efficient manner.

• AMD Resource Manager: Helps organizations control and optimize how users and teams access GPUs, data, and compute resources. It improves GPU utilization through fair scheduling and shared access, while offering dashboards to monitor usage across projects and departments.

• Kubernetes AI Workload Orchestrator (Kaiwo): Enhances GPU efficiency by reducing idle time through intelligent scheduling. It manages AI job placement using a Kubernetes operator and supports features like multiple queues, fair sharing, quotas, and topology-aware scheduling to run workloads more effectively.

• Kubernetes Platform: Serves as the core container orchestration layer that powers the deployment, scaling, and management of AI workloads. It provides the flexibility and reliability needed for tasks ranging from training large models to running production inference.

• Cluster Forge: Simplifies the setup of a production-ready AI platform by automating the deployment of Kubernetes control and compute planes. It integrates open-source tools and packaged AI workloads, enabling teams using AMD hardware to get started within hours.

• AMD Inference Microservices (AIMs): Streamlines the process of serving AI and LLM models by automatically selecting optimal runtime settings based on the model, hardware, and user inputs. Its expanding catalog of prebuilt microservices makes deploying inference workloads fast and efficient.

Deploy AMD Enterprise AI Platform

In this section, you deploy the AMD Enterprise AI Platform using the bloom installer. You download and prepare the Bloom binary, configure the required YAML settings, and launch the browser-based installation interface over a secure SSH tunnel. After reviewing and confirming the configuration inside the installer UI, you trigger the full platform deployment.

1. Download the official bloom binary. Visit the GitHub releases page to get the latest version.

   console Copy

   $ wget https://github.com/silogen/cluster-bloom/releases/download/v1.2.2/bloom
   

   Explain Code

2. Make the bloom binary executable.

   console Copy

   $ chmod +x bloom
   

   Explain Code

3. Create the Bloom configuration file.

   console Copy

   $ nano bloom.yaml
   

   Explain Code

4. Add the following content to the file. Replace amd-ai-suite.example.com with your domain and /dev/vdb1 with the block device attached to your server. Visit the ClusterForge GitHub release page to download the latest release package.

   yaml Copy

   DOMAIN: amd-ai-suite.example.com
   OIDC_URL: https://kc.amd-ai-suite.example.com/realms/airm
   FIRST_NODE: true
   GPU_NODE: true
   CERT_OPTION: generate
   USE_CERT_MANAGER: true
   CLUSTER_DISKS: /dev/vdb1
   CLUSTERFORGE_RELEASE: https://github.com/silogen/cluster-forge/releases/download/v1.5.2/release-enterprise-ai-v1.5.2.tar.gz
   NO_DISKS_FOR_CLUSTER: false
   

   Explain Code

   In the above configuration:

   • DOMAIN: Sets the base domain the platform uses.
   • OIDC_URL: Defines the Keycloak authentication URL for the airm realm.
   • FIRST_NODE: Marks this server as the initial node in the cluster.
   • GPU_NODE: Enables GPU capabilities for this node.
   • CERT_OPTION: Defines how certificates are created (generate = auto-generate).
   • USE_CERT_MANAGER: Enables Cert-Manager for managing TLS certificates.
   • CLUSTER_DISKS: The disk or partition the cluster uses for storage.
   • CLUSTERFORGE_RELEASE: URL of the ClusterForge package required for installation.
   • NO_DISKS_FOR_CLUSTER: Indicates whether the cluster should run without disks (false = use the disk listed above).

5. Start the installation process.

   console Copy

   $ sudo ./bloom --config bloom.yaml
   

   Explain Code

   This command launches a local installation interface at http://127.0.0.1:62078.

6. Open an SSH tunnel to access the installation UI.

   console Copy

   $ ssh -L 62078:127.0.0.1:62078 USERNAME@SERVER-IP
   

   Explain Code

   Replace USERNAME with your server username and SERVER-IP with your server's public IP.

7. Open the installation UI in your browser.

   http://localhost:62078

8. Follow the instructions in the web interface and review any configuration options that require changes.

9. After you finalize the configuration, click Generate Configuration & Start Installation to begin the deployment.

   Note

   The deployment usually takes 20 minutes to finish.

Configure SSL and Access the Resource Manager UI

In this section, you configure a wildcard Let's Encrypt SSL certificate so you can access the Resource Manager UI securely. The UI enforces HTTPS, and without a valid TLS certificate, you cannot open the interface.

You can use either of the following SSL methods:

• A Let's Encrypt wildcard SSL certificate for your domain.

• A SAN-based SSL certificate that includes these subdomains:

  • airmui.amd-ai-suite.example.com
  • airmapi.amd-ai-suite.example.com
  • argocd.amd-ai-suite.example.com
  • gitea.amd-ai-suite.example.com
  • kc.amd-ai-suite.example.com
  • longhorn.amd-ai-suite.example.com
  • minio.amd-ai-suite.example.com
  • openbao.amd-ai-suite.example.com

  Replace amd-ai-suite.example.com with the domain or subdomain you selected for your installation.

Generate a Wildcard Let's Encrypt SSL Certificate

In this section, you generate a Let's Encrypt wildcard SSL certificate for your domain using certbot.

1. Update your package index and install certbot.

   console Copy

   $ sudo apt update && sudo apt install certbot -y
   

   Explain Code

2. Generate a wildcard certificate.

   console Copy

   $ sudo certbot certonly --manual --preferred-challenges dns -d '*.amd-ai-suite.example.com'
   

   Explain Code
   Note

   Creating a wildcard SSL certificate requires domain ownership verification. Certbot can automate this process using DNS plugins, but only for supported DNS providers. The --manual method works with any DNS provider, but requires you to create a TXT record manually.

   Certbot displays output similar to:

   Please deploy a DNS TXT record under the name:
   
   _acme-challenge.amd-ai-suite.example.com.
   
   with the following value:
   
   zyuf8RXatvvwgPFH-gqj.......................

   From the output, copy the record name and value, then open your DNS panel and create a TXT record using those values.

3. After the DNS record propagates, press ENTER to continue domain validation.

   Output:

   Successfully received certificate.
   Certificate is saved at: /etc/letsencrypt/live/amd-ai-suite.example.com/fullchain.pem
   Key is saved at:         /etc/letsencrypt/live/amd-ai-suite.example.com/privkey.pem

   From the output, note the certificate paths for the later use.

Create a TLS Secret and Access the Resource Manager

In this section, you create a Kubernetes TLS secret that the HTTPS gateway uses to serve your wildcard certificate. After applying the certificate and restarting the gateway, you access the AMD Enterprise AI Resource Manager UI and sign in using the default credentials to complete the initial login process.

1. Create the TLS secret.

   console Copy

   $ kubectl create secret tls cluster-tls \
       -n kgateway-system \
       --key /etc/letsencrypt/live/amd-ai-suite.example.com/privkey.pem \
       --cert /etc/letsencrypt/live/amd-ai-suite.example.com/fullchain.pem
   

   Explain Code
   Note

   Ensure your user has permission to read the certificate files. If not, prepend sudo to the command. If you run the command with sudo, ensure the root user has access to Kubernetes credentials (via /root/.kube/config or system-wide configuration).

2. Verify that Kubernetes created the secret successfully.

   console Copy

   $ kubectl get secret/cluster-tls -n kgateway-system
   

   Explain Code

   You should see the cluster-tls secret listed in the output.

3. Restart the HTTPS gateway so it loads the new certificate.

   console Copy

   $ kubectl rollout restart deployment/https -n kgateway-system
   

   Explain Code

   After the restart completes, access the AMD Enterprise AI Resource Manager dashboard using the url below:

   https://airmui.amd-ai-suite.example.com

4. Click Sign in with Keycloak. It redirects you to the Keycloak login page.

   

5. Enter the following default credentials. Replace amd-ai-suite.example.com with the domain you configured.

   • Username: devuser@amd-ai-suite.example.com
   • Password: password

   After you log in, the system prompts you to reset the default password. Enter a strong alphanumeric password twice to secure your account.

6. After you reset the password, the dashboard loads and displays the AMD Resource Manager interface.

   

Key Feature of the Platform

• Optimized GPU utilization and lower operational costs: Intelligent scheduling maximizes GPU usage, reduces waste, and lowers overall compute costs.
• Unified AI infrastructure: Brings all AI tools and environments together into a single, consistent platform for easier collaboration and governance.
• Accelerated Time-to-Production: Built-in microservices and streamlined workflows help teams move AI models into production faster.
• AI-native workload orchestration: Purpose-built scheduling and inference services ensure efficient, high-performance execution of AI workloads on AMD Instinct™ GPUs.

Conclusion

By following this guide, you deployed the AMD Enterprise AI Platform on Vultr using AMD Instinct™ GPUs and the Bloom installer. You also explored the platform's core components AI Workbench, Resource Manager, Kaiwo, Cluster Forge, and AIMs and learned how they work together to deliver a unified, scalable, and high-performance AI infrastructure.