Implementing a Transputer system using RasPi Pico and RAM expansion
trochilidae.blogspot.com
trochilidae.blogspot.com
Atari released the ATW in 1987, but it was a flop. I think the main reason was unfamiliarity- a parallel cpu architecture running unix-like OS but with no mmu, and no grandstand applications sunk it.
Had the transputer launched as a coprocessor (even the T212 was 10x faster than the 68k fpu) it would have allowed developers to get familiar with OCCAM etc. Alongside a more traditional arch before launching a 32bit machine with the full TRAM style modules.
Interestingly, this might well have changed the computing landscape, using transputers as Xeon Phi style GPUs before GPUs were a thing.
It would have also led to an interesting situation when Motorola ended the 68k line. Rather than going to power PC as the Mac and Amiga did, the ST/TT could have gone full transputer instead.
Lastly, the transputers are quite small (20k gates) and power efficient. ARM might not have had the same legacy with a competitive transputer scene as a result of using it in volume as a coprocessor.
This is a great project, with lots of enthusiastic people in the comments section, very pleased to see a revival of the concepts!
OCCAM was lovely, in an academic sense, but it was effectively a different evolutionary branch of computing. That branch has perhaps more influence today than ever, with Go's use of CSP-like channels, but OCCAM never really took off.
Inmos sold TDS (the IDE/SDK) for a lot of money just when Open Source was really starting to kick off. The folding editor, while it had its good points, was also strange from a mainstream point of view, adding to the barrier to entry.
By the time you could get common languages able to compile for the Transputer, the T425/T800 were outpaced by the rest of the industry, and the T9000 only really approached being available as Inmos collapsed. And by that time, it was already obsolete anyway.
But XMOS has done fairly well, albeit in a limited niche.
Very far from the first or last time that happened. :-(
I think ARM2 would have been a very competitive choice in that era.
ARM1 had 24800 transistors or ~6000 gates [0]. ARM2 had 27k transistors [1], couldn't find a source about how many gates it got, but probably less than 7000. No FPU, though, but up to 16 cycle integer multiplier and multiply-add. In comparison, 16 bit x 16 bit = 32 bit multiply on an MC68000 could take up to 70 cycles.
Nevertheless, transputers were pretty cool way back then. I remember drooling after them. As is the project in the linked article.
(Side note: As much as I like Motorola 68000 CPU range, it's a pretty sad comparison — it had 68000 transistors and a fraction of performance of an ARM2 CPU.)
[0]: https://en.wikichip.org/wiki/acorn/microarchitectures/arm1
[1]: https://en.wikichip.org/wiki/acorn/microarchitectures/arm2
(!) Is that the origin of the chip's name?
The ARM2 was highly competitive, yes. I still own my original ARM2 system from the late 1980s, an Acorn Archimedes A310.
It has been odd to see the techie world getting excited over Apple's Arm-ISA CPU performance, some 35 years after much the same thing happened when the ARM2 hit the retail market.
I bought my A310 used for £800 in 1989. At the time, my employer's top-end fastest system was an IBM PS/2 Model 70 H21 (I hope I have the model number right): an 80386DX at 25MHz with _external cache_, a radical new feature.
It cost about £10,000 (without optional extras such as a keyboard or MS-DOS).
My home computer was 2-4x faster, at about 10% of the price.
I had a software PC emulator that emulated a ~2.5MHz 8086: slow, but usable for DOS. The underlying OS meant that disk access was super fast. In other words, an ARM2 could emulate an x86 in software at competitive speeds, as an Apple M1 can do now versus modern Intel CPUs.
I think the snags with it were that it had 32-bit wide buses, not long after Motorola introduced the 68008 _narrowing_ the chip's buses to 8 bits wide to reduce costs, and a couple of years before Intel introduced the 80386SX in 1988, narrowing the 80386DX's 32-bit buses to 16 bits, again for economical purposes.
A full 32-bit wide motherboard was expensive.
The ARM2 really needed its original chipset, too, and that too had restrictions, e.g. MEMC imposing a large-for-the-time 4 kB page size.
Later when PCs with an ISA bus became less common we developed a USB interface to the device. Still needed to source the C011 transceiver chips but we got those free issue from the instrument manufacturer.
It was not friendly to users who wanted a free choice of languages.
The fact that some languages (and OSes back then) were not well-suited to parallel processing, and thus programming parallel programs, which you needed to make effective use of transputer systems, did not get across to people.
Now, I suspect most people think we have a very wide, free choice of programming languages -- but the truth is that for 25+ years, mainstream CPU designs are C-centric, designed to be easily programmable in C and to run OSes implemented in C.
This wasn't true in the 1960s-1980s, when there were, for example, many 9-bit, 12-bit, 18-bit, 24-bit and 36-bit machines, on which C did not sit well at all.
The transputer arguably came in a bit too early for widespread adoption of parallel multicore computers, but also was paradoxically too late, as it needed non-C-like languages and some of its own OSes for most effective use.
Occam polarised people. Some people liked it, but many hated it.
Interesting, the transputer OS Helios later got ported to other architectures: https://github.com/axelmuhr/Helios-NG