Four machines, one system
A Mac that builds, a laptop that draws and trains, a $23 server that answers, and a GPU that exists only while it is called. The topology is the architecture.
Labels. [MEASURED] — from a runbook, a snapshot or a receipt in the pack. [FACT] — a file you can check. [OPINION] — mine.
The system has no single computer. It has four, and the interesting thing is that none of them is redundant:
| Machine | What it is | What it does | What it costs |
|---|---|---|---|
| Mac | a laptop, arm64 | writes and builds the publication, holds the git source of record, runs the deploy | sunk — it is the machine I work on anyway |
| M2 | i7 / 64 GB / RTX 3080 Ti / 4 TB, Fedora | the GIMP print studio, local GPU experiments, offline ingestion; hot standby during the VPS move | electricity, and it is already owned |
| VPS | OVH VPS-4, 8 vCores / 24 GB / 200 GB NVMe | the engine API, doc-nor, the shop, the tarot tool, the publication, backups | $23.37/month |
| Modal | L40S 48 GB, scale-to-zero | the 35B reader and the FLUX image worker | $0 idle, ~$2.40–2.60/hour awake |
Four machines, four different billing models, one system. This is the piece about why that is the right shape — and what breaks when you pretend it is one computer.
Each machine has a different job because each job has a different duty cycle
The mistake a solo operator makes is buying one machine that is good at everything. The bill for that is a GPU that idles all day, a CPU that is over-specified for serving, and a single point of failure that is also the development environment.
This piece is part of the members' archive — $8 a month or $80 a year.
$8 / month · $80 / year
Sign in if you are already a member, or create an account.