CJW–07.1
roles, not random boxes
personal lab
guide
V.0.5.0
compute
compute is organized by job. the kubernetes proof cluster runs as talos virtual machines on a proxmox host. other machines cover gpu inference, development, and supporting services — each with a reason to exist.
01
kubernetes proof host
a proxmox ve host runs the primary cluster as virtual machines — three control-plane and three worker talos nodes — so the platform can be rebuilt from infrastructure-as-code instead of snowflake installs.
- —proxmox ve
- —6 talos linux vms (3 control / 3 workers)
- —terraform-managed node lifecycle
- —zfs-backed vm storage on the host
02
supporting compute
additional compact and server-class systems in the broader lab cover utility services, experiments, and denser workloads when the proof cluster is not the right place to run them.
- —utility mini nodes
- —high-memory / expansion paths
- —role-based placement
03
gpu path
an rtx 3090-class card (24 gb vram) supports local llm inference, embeddings, agents, and cuda work — kept as a deliberate path, not casually bolted onto every cluster node.
04
workstation
apple silicon workstation for development, automation, media workflows, and ai experimentation against the lab.
- —m1 max class
- —unified memory
- —local tooling