If you liked Transputers, you might want to also read about Adapteva and their Epiphany core for a more recent attempt at something similar-ish.
I still have two of their prototype machines from their Kickstarter - two ARM cores to run Linux, with an Epiphany chip with 16 cores in a 4x4 grid. But their goal was scaling it up to 64 cores or up to I think 4K cores on a board. Each core had a small amount of on core RAM and four buses to each side in the grid, and you could access the memory of every other core with a predictable latency (one cycle per "hop"), so if you planned things carefully, you could have them working in lockstep.
It's an interesting space, but hard because the first difficult question you need to answer - which strips away a whole lot of potential use-cases and many of the most profitable one - is "why not a GPU?".
Just before Epiphany, there was also the Tilera, which had a lot in common with the Transputer. Our lab got one and we played around with it, but it was a pain to program. Transputer had OCCAM, Tilera chased after the C model and shared coherent memory.
The Tilera TILE architecture lives on in NVIDIA's DPU.
I still have two of their prototype machines from their Kickstarter - two ARM cores to run Linux, with an Epiphany chip with 16 cores in a 4x4 grid. But their goal was scaling it up to 64 cores or up to I think 4K cores on a board. Each core had a small amount of on core RAM and four buses to each side in the grid, and you could access the memory of every other core with a predictable latency (one cycle per "hop"), so if you planned things carefully, you could have them working in lockstep.
It's an interesting space, but hard because the first difficult question you need to answer - which strips away a whole lot of potential use-cases and many of the most profitable one - is "why not a GPU?".