Thread by @LuigiThirty

The most fascinating thing about the transputer tech is that they were *right* but not targeting the right niche

they wanted them to be a universal general-purpose computer system but that's not really where you need massively-parallel systems

even at the time they caught on for image processing and scientific experiments, places where you'd use compute units today

my video card has 36 shader processors that contain thousands of processors running in parallel to draw 2B's ass

The biggest problem was Moore’s Law held way longer than expected and you couldn’t scale transputer designs to compete against early-90s RISC systems or even commodity hardware, and that they could never make them cheap enough to compete against microcontrollers

Their idea was “if every device in your system contains a transputer, every computer becomes a massively parallel supercomputer” since every device controller could contribute its spare cycles to running general-purpose operations

that… never really happened, obviously

turns out you don’t need a massively parallel supercomputer unless you’re doing very specific things like network operations or digital imaging

ISDN hardware saw a lot of transputers being used for example

analog signal analyzers too until DSPs outclassed them

and what is CUDA but running a program on a gazillion compute units in parallel. it was GPGPU before we even had GPUs!

I’ve got a tabletop Tektronix SS7 analyzer here that features a whopping 7 transputer cores on two enormous cards, used for analyzing telephone system traffic