[Instantix]

I didn't know that transputer were made for the Amiga. Some years ago I did some research about the Atari one and the conclusion was: inefficacy. The serial link linking each transputer being a bottleneck. And more transputers you add, more complexity in tasks distribution you add so the performance curve was flattening.

I'm curious to know if multi-core processors were designed with this lesson in mind.

[justinlloyd]

The Atari Transputer system was unfortunately hampered by a number of design factors. Transputers did have a bottleneck with inter-chip communication, but generally you programmed to the transputer architecture, rather than a general purpose computer architecture. This meant you broke up your data in to chunks, and then let each transputer do its own thing, and then regroup at the end, perhaps using a separate transputer to reassemble the data, much like you would with a modern GPU. You could treat your functions either like a pipeline or like a matrix, but you had to create the algorithm to work with that architecture. Part of my job was showing companies who were deploying transputer based systems how to port their work to this architecture. That said, the transputer to transputer communications bus was pretty optimal, for the time, compared to how most multiprocessor computers handled their work.