AI ML GenAI on Data Center-Class GPUs with Red Hat OpenShift

In this lesson you’ll watch our entire lab come to life—from powering-on the jump host to mapping the network VLANs that tie every component together. We start at the hypervisor console, allocate CPU, RAM, and storage for each VM, then flip over to the bare-metal H100 node for a quick firmware sanity-check. Next, you’ll see how we:

1. Attach the virtual ISO repository and upload the Agent image.

2. Configure a dedicated VM network on vSwitch1 and test end-to-end connectivity with ping and dig.

3. Set static IPs for the API VIP, Ingress VIP, masters, and worker nodes.

4. Install and start dnsmasq so our cluster names resolve instantly.

5. Snapshot the environment—giving us a clean rollback point before the actual OpenShift install.

By the end you will be able to

• Recreate the entire VM topology on any ESXi or KVM host.

• Verify base networking without touching a single OpenShift command.

• Capture a “golden snapshot” so you can experiment fearlessly in later lessons.

Prerequisites
A fresh hypervisor, the jump-host ISO, and about 30 GB of free SSD space.

In this lesson we’ll take a raw bare-metal host and turn it into a battle-ready VMware ESXi 8.0 U2 hypervisor—your launchpad for the OpenShift control-plane VMs. You’ll watch every step on screen, from creating bootable media to the first login in the vSphere HTML5 client. Along the way we will:

1. Validate firmware compatibility with the VMware Hardware Compatibility Guide.

2. Burn the ESXi ISO to USB (or mount it via iDRAC virtual media) and walk through the purple-screen installer.

3. Tune critical BIOS settings: UEFI, VT-x/AMD-v, SR-IOV, and PCIe Gen 5 lanes.

4. Configure the Direct Console UI (DCUI) for static IP, DNS, and NTP.

5. Activate SSH, upload license keys, and patch to the latest build via Lifecycle Manager.

6. Create datastore1 on NVMe, build vSwitch1, and carve a dedicated VM Network port-group for OpenShift traffic.

By the end you will be able to

• Install ESXi 8.x on virtually any supported server in under 10 minutes.

• Apply best-practice security (strong root password, NTP, key-based SSH).

• Provision storage and networking exactly as the Agent-based Installer expects.

• Capture a clean “hypervisor-baseline” snapshot so you can redeploy VMs without reinstalling ESXi.

Prerequisites
• A blank server with at least one SSD/NVMe drive
• ESXi 8.x ISO + license or evaluation key
• Access to server iDRAC/iLO or a bootable USB port

In this lesson we turn a blank VM into the jump-host—your one-stop command center for every OpenShift, DNS, and automation task in the course. You’ll see the full, click-by-click walk-through:

1. Mount the RHEL 9.4 ISO and choose the Minimal Install profile (no GUI bloat).

2. Partition the disk: 30 GB for /, the rest for /var, XFS for resilience.

3. Set static networking to 192.168.1.10/24 and hostname jump-01.ocp.lab.local.

4. Register with Red Hat Subscription Manager, enable BaseOS/AppStream, and run dnf update -y.

5. Install core packages: podman, git, jq, dnsmasq, httpd-tools, chrony.

6. Generate an ED25519 SSH key pair for later use in install-config.yaml.

7. Harden the box: enforce SELinux, open only SSH in firewalld, enable chronyd.

8. Create the ~/ocp-install workspace and drop in the Red Hat pull secret.

9. Add bash completion for oc, customise the prompt, and snapshot the VM.

By the end you will be able to

• Stand up a lean, secure RHEL jump-host in under ten minutes.

• Equip it with every tool—openshift-install, oc, yq, helm—needed for the rest of the lab.

• Validate connectivity and DNS resolution before generating the Agent ISO.

Prerequisites

• ESXi/kvm VM with 4 vCPU, 8 GB RAM, 50 GB disk

• RHEL 9 ISO (or Rocky/CentOS Stream 9 if subscriptions aren’t available)

• Internet access for initial dnf update (or local repo if air-gapped)

In this lesson we move from planning to execution: you’ll watch three master VMs boot from the custom agent.x86_64.iso, self-register with the Assisted Service, and converge into a healthy, highly available OpenShift 4.18 control plane. The video captures every screen—BIOS splash to oc get nodes—so you can follow at your own pace.

Key milestones you’ll see on-screen

1. Mount & power-on master0, master1, master2 in ESXi with the ISO attached.

2. Live log tail of openshift-install agent wait-for bootstrap-complete, decoding each status: discovering → known → installing → done.

3. master0 established as the rendezvous host, running the embedded Assisted Service.

4. Automatic ignition of control-plane services—etcd, kube-api, scheduler—verified via oc get pods -n openshift-kube-system.

5. First successful login with the kubeadmin credentials and a tour of the web console.

6. Final health sweep: ClusterVersion Available, all Operators green, and the control-plane marked Ready.

7. Snapshotting the masters for a clean rollback point.

By the end you will be able to

• Generate and boot an Agent ISO that discovers hosts with zero external provisioning network.

• Interpret bootstrap logs to diagnose slow or failed installations.

• Validate a brand-new control plane using both CLI (oc) and the web console.

• Capture a “post-control-plane” snapshot—your safe reset point before adding GPU and CPU workers.

Prerequisites

• Lesson 5 jump-host with openshift-install and oc configured.

• Master VMs created from the template (4 CPU / 16 GB / 120 GB each).

• DNS, VIPs, and load balancer already in place (covered in Lesson 4).

Before a single Kubernetes pod can claim GPU time, we need to guarantee the hardware is healthy and perfectly tuned. In this 11-minute deep-dive you’ll ride along as we log into Dell iDRAC, audit every sensor, and apply the five BIOS tweaks that keep a Hopper-class GPU happy under OpenShift. You’ll see:

1. System → Overview – verifying service tag, firmware dates, and BIOS 2.13.0+ for PCIe Gen 5 throughput.

2. Maintenance → Firmware Inventory – confirming H100 firmware ≥ HBM3 95 and queueing updates in a single reboot job.

3. Configuration → BIOS Settings – enabling UEFI, SR-IOV, and setting System Profile = Performance with Energy-Efficient Turbo.

4. Dashboard → Sensors – live look at GPU temperature (< 35 °C idle), fan RPM, inlet temp, and input power.

5. SSH RACADM session – grabbing the GPU UUID, dumping hwinventory, and exporting a SupportAssist log bundle for your records.

6. Optional toggle: Persistent GPU UUID for future VMware migrations.

By the end you will be able to

• Verify and upgrade H100 firmware entirely through iDRAC—no OS boot required.

• Apply best-practice BIOS and thermal settings that prevent driver crashes and throttling.

• Collect sensor data and log bundles for proactive support or root-cause analysis.

• Capture a “pre-cluster” hardware baseline snapshot you can compare against months later.

Prerequisites

• Physical access or iDRAC credentials for the NVIDIA H100 server.

• Firmware packages downloaded from Dell Support (if upgrades are needed).

• SSH client on the jump-host to run racadm.

In this lesson, we wire up the arteries that let every OpenShift component talk—cleanly, securely, and with zero surprises at bootstrap. You’ll watch us:

1. Design the CIDR map – carving a /24 for cluster traffic and reserving IPs for the API VIP (6443), Ingress VIP (443/80), masters, and workers.

2. Stand-up a Layer-4 HAProxy on the jump-host, forwarding raw TCP for the API and Machine Config Server plus HTTPS/HTTP for Ingress.

3. Spin up dnsmasq with authoritative A & PTR records—including the wildcard *.apps.<cluster>.<domain>—and validate resolution with dig.

4. Add a static route on the GPU node to ensure east-west traffic stays on the 10 GbE LAN.

5. Test port reachability with nc -zv <vip> 6443, confirming the load balancer is live before masters ever boot.

6. Capture a tcpdump trace to show the first SYN packets hitting all three control-plane nodes—proof that fail-over is instantaneous.

By the end you will be able to

• Allocate VIPs and node IPs that satisfy the Agent-based Installer’s strict requirements.

• Configure dnsmasq (or any DNS engine) with the exact record set OpenShift needs.

• Deploy a lightweight HAProxy config that survives reboots and requires no Layer-7 smarts.

• Validate the whole stack—DNS, VIP, and ports—in under two minutes with command-line tools.

Prerequisites

• Jump-host from Lesson 5 with root access.

• IP plan (spreadsheet or whiteboard) for your lab network.

• Basic understanding of firewalls; outbound 443 allowed for RHEL updates.

In this lesson the capstone of the installation arc: you’ll watch the custom agent.x86_64.iso spin up our entire cluster—from bootstrap ignition to a fully operational control plane plus one CPU-only worker. Every stage of the Agent-based workflow is captured live on screen, giving you a crystal-clear blueprint you can replay in your own lab.

Milestones you’ll see in real time

1. Boot & Discovery Each VM mounts the ISO, sends its inventory, and flips from discovering → known in the Assisted UI.

2. Validation Phase The installer checks CPU, RAM, disks, and network—see how to fix “pending-input” hosts fast.

3. Bootstrap Event master0 (the rendezvous host) launches the temporary control plane; logs scroll inside openshift-install agent wait-for.

4. Etcd Quorum Masters pivot to “installing-in-progress,” write RHCOS to disk, and reboot into real control-plane services.

5. Agent ISO Auto-joins the Worker Watch the CPU worker enrol itself and register with the kube-API—no manual CSR needed.

6. First Console Login Use the freshly generated kubeadmin password, check ClusterVersion, and tour the Operator status dashboard.

7. Post-Install Snapshots Learn exactly when—and why—to snapshot your VMs for an easy rollback point.

By the end you will be able to

• Run the Agent ISO from power-on to “Install complete!” with zero manual intervention.

• Decode installer log messages, spot bottlenecks, and restart failed hosts safely.

• Validate cluster health via both oc CLI and the web console before moving to GPU integration.

Prerequisites

• Network, DNS, and HAProxy configured (Lesson 8).

• install-config.yaml and agent-config.yaml in ~/ocp-install.

• Jump-host with openshift-install and oc binaries.