Part 36: The 128 GB Workstation📋

"128 GB of DDR5 is €500. The freelancer who bills four-figure rates pays for it on the first day of the second client."

Why📋

Part 02 made the case for 64 GB as the sweet spot for one client with a real HA cluster. This part is about the next sweet spot: 128 GB, the freelancer with multiple clients.

The thesis: 128 GB of DDR5 RAM in a consumer workstation is achievable, affordable, and exactly the right amount for two or three k8s clients running in parallel. The math works. The investment pays back inside a month.

What 128 GB buys you📋

The realistic ceiling on 128 GB is three clusters if at least two of them are not HA. Two HA clusters is comfortable. Three HA clusters does not fit.

The "host + IDE" budget is ~16 GB:

• Operating system: ~4 GB
• VS Code with .NET / Java extensions, language servers: ~3 GB
• Browser with 30 tabs: ~5 GB
• Slack + Zoom + 1Password + Spotify: ~4 GB

This is conservative. Power users (Rider open + Postman + DBeaver + DataGrip + a YouTube tab) consume more. The 128 GB box is sized for "comfortably handles three clients", not "every possible host application".

The hardware📋

A 128 GB DDR5 workstation in mid-2025:

Plus the rest of the build (case, PSU, cooler, GPU if needed): another €500. Total rig: ~€1,800 for a workstation that hosts three real Kubernetes clusters in parallel.

For a freelancer billing €600-800 per day, this is two days of work. The investment pays back the first time it lets you bill two clients in parallel without their work interfering — usually within the first month.

The math against real workloads📋

The 80 GB used by "2× k8s-multi" breaks down as:

That leaves 55 GB of headroom on a 128 GB workstation. The headroom absorbs:

• Build cache (~5 GB during a dotnet build or mvn package)
• Container image pulls (~3 GB at peak)
• Browser memory spikes
• The third client when it shows up

The math is intentionally conservative. A real freelancer hits the limits less often than this analysis suggests because not every cluster runs every workload simultaneously — the dev cluster might be idle while the stage cluster runs a test.

When you hit the wall📋

The wall is "I cannot fit a fourth client". When that happens, the answers are:

1. Drop a client to k8s-single. A 16 GB k8s-single replaces a 32 GB k8s-multi for clients where the dev environment does not need multi-node fidelity. Frees ~16 GB for a new client.
2. Spin clusters down when not in use. homelab vos halt acme drops Acme from RAM completely (the VMs are stopped, the disks are kept). When you context-switch back, homelab vos up acme brings them back in ~60 seconds. The state survives the halt because the volumes are persistent.
3. Move a client to a second machine. The multi-host scheduling pattern from homelab-docker Part 46 lets you spread VMs across two physical machines. A second laptop, even a modest one, is cheaper than a new RAM upgrade.
4. Buy more RAM. Some Threadripper / Xeon workstations support 256 GB. The math then is "five clients in parallel". For the freelancer who consistently has more than three concurrent clients, this is the next step.

Cost vs the alternatives📋

The traditional answer to "I have multiple clients" is a separate machine per client. A modest €700 laptop per client. Three clients = €2,100 in laptops. Plus the overhead of context-switching machines (different keyboards, different fonts, different ssh agents, different browser bookmarks).

A 128 GB workstation at €1,800 replaces three separate machines and removes the context-switching tax. The pay-back is fast and the daily quality-of-life improvement is measurable.

What this gives you📋

A clear answer to "what hardware do I need". 128 GB. Why. What it costs. What it buys. When you hit the wall. The freelancer can plan their hardware budget against this and not be surprised.

Cross-links📋

• Part 02: The 64 GB Sweet Spot
• Part 03: Multi-Client Isolation
• Part 25: Topology Comparison
• Part 39: The Pro Consultant Workflow
• HomeLab Docker — Part 47: Cost Tracking