File 002 · Local Intelligence
Record 030 · Machines · 2022
22.30.BMac Studio
Apple
“The small Mac that made 64–192GB unified memory a local-LLM workstation.”
A lot of RAM, no GPU slot.
M1/M2/M4 Ultra Studios run 70B+ in memory that CPU and GPU share. No 4090, no blower, a fan you can live with. The quiet local rack.
Opened high-memory local inference without a datacenter GPU.
Filed notes
- Unified memory at workstation sizes
- Quiet always-on inference
- mlx/llama.cpp Metal
SIC
3571
Electronic Computers
NAICS
334111
Electronic Computer Manufacturing
