Atlas record: https://www.aitimeline.dev/en/events/evt-0148/
Research · 1985
The Connection Machine
Danny Hillis designed a computer of tens of thousands of simple processors, each with its own memory, joined in a hypercube.
Why it matters
The bet on massive parallelism instead of one faster processor reached a production machine for the first time.
The idea came directly from AI problems: search in a semantic network and pattern recognition both fall naturally into thousands of identical small operations. The CM-1 held 65,536 one-bit processors. Thinking Machines did not survive commercially, but the thesis that many simple units beat one complex one was borne out, fifteen years later and on graphics processors.
Event record
- Event date
- 1985
- Timeline date
- Event date
- Verification
- Sources gathered automatically · September 17, 2026
- Lines
- ID
- evt-0148
Year of the MIT thesis; Thinking Machines shipped the CM-1 the same year.
Related events
Earlier
- Related Japan's Fifth Generation project 1981
A second bet of those years on parallel architecture for AI workloads.
- Related Kung and Leiserson's systolic arrays 1978links to this record
Both answer the limit of a single von Neumann machine with many simple processors, but a systolic array pumps data rhythmically through a grid, while the Connection Machine joins its processors in a hypercube.
- Related Lisp 1960
The machine was built for parallel data processing rather than sequential execution of symbolic code.
Later