The company I worked for in the early 90s decided to use Transputers for their custom embedded systems, so I had the privilege of working with them in development and production environments for a few years.

They were an absolute joy to work with. You would structure your code into tasks that could be performed in parallel, with a simple queue for inter-task communication. You could then debug the code on a single transputer (residing on an I/O card plugged into your computer).

The production board typically contained several transputers, all connected through serial links, and with one off-board link connector. To upload your code to production, you connected a link from the transputer on your dev card to the production board, placed the board in "boot from link" mode and downloaded the code to the target. A "placement" file directed where the various tasks should run, so the tasks which ran on a single processor in dev, were loaded into the specified processors on the target. I.e. going from single processor to multiprocessor execution without even requiring a re-compile.

The T800 was the fastest floating-point processor available at the time and blew our previous generation processors out of the water. Everyone was sad when INMOS /SGS Thompson faltered with the development of the T9000 and lost their performance edge, eventually resulting in the demise of the platform.