EXPERIMENT 002 / PHYSICAL COMPUTE
MICROBLADE.
A compact compute system where the chassis, airflow, wiring, and cluster all evolve together instead of being treated as separate problems.
SYSTEM PROFILE / 01
Compute that can grow without losing its shape.
OBJECTIVE / 02
Make infrastructure feel like a product, not a pile of parts.
MicroBlade is a 3D-printed rack built around compact t740-class x86 nodes. The goal is a physical system that stays clean as it expands: predictable mounting, usable airflow, deliberate cable paths, and enough access to service a node without dismantling the entire cluster.
The interesting part is the boundary between mechanical design and software. Better structure makes networking, orchestration, cooling, and future upgrades easier to reason about—and makes the cluster something that can be iterated instead of merely assembled.
QUESTIONS / 03
What the rack is really testing.
Can small compute scale cleanly?
The system is being designed so capacity can grow one node at a time without turning the rack, wiring, or cooling into a rebuild every time another machine is added.
What should the physical system solve first?
Mounting, airflow, networking, cable routing, and access all matter before orchestration does. The hardware should make iteration easier instead of becoming the thing that slows it down.
How modular can the rack become?
The current design leaves room for the surrounding system to evolve: switching, power management, cooling, and new compute modules can become first-class parts instead of afterthoughts.
WORKING LOG / 04
The chassis is part of the experiment.
This field note will grow with the hardware. Fit tests, thermal observations, rack revisions, networking choices, and node behavior can be added as they become real measurements instead of assumptions.
END OF CURRENT SIGNAL