*reads up about transputers*

...I never even HEARD of the Occam programming language before... :o

...seems like the biggest hurdle towards these systems was cost vs. performance, plus one thing I do know for sure is that not a lot of programmers understood the concept of multi-threaded programming. Even nowadays coders aren't typically all that good at it unless they're highly experienced.

The Sega Saturn is a good example of this. There was a great debate back in the mid to late 90s as to which was more powerful, the Saturn or the Playstation. Theoretically, the Saturn can outperform the Playstation, but there's a catch. The Saturn has TWO processors, the Playstation only has one. Any game written to only use one processor on the Saturn will run worse on the Saturn than on the Playstation, whereas any game written to use both processors equally will run better on the Saturn than on the Playstation.

Many of the companies releasing games on both systems would code for the Playstation first, then do a hack-job port to the Saturn that doesn't really utilize its second CPU, thus many games which showed up for both platforms ran better on the Playstation, even though if they were coded better, they'd run better on the Saturn. :P

My point is that massive parallelism is great for performance, IF the person writing the code knows how to take advantage of it, and programmers who can are the minority, so I imagine this is a part of what led to the transputer not being widely adopted.

Good explanation. I would guess that a lack of marketing, combined with a prohibitively high cost and lack of software would factor in as well, like for many other failed platforms.

The Atari name itself killed a lot of promising projects back then. They were known as a "game company" and who wants to buy a serious computer for business or scientific application from a "game company"?

Atari was so good at marketing that if they bought KFC they'd market it as "Warm dead bird"

There were transputer boards for amigas and PCs as well.

At the time [late 80s?] I thought it heralded a new world of high speed applications, I suspect lack of imagination amongst potential users [as well as amongst potential suppliers/vendors] also played a part in the lack of takeup. I remember bringing transputers up at a Government conference on high performance computing in statistics back then... just got blank looks, nobody had even heard of it despite good press in computing mags.

More Transputer (specifically the PC cards) here: http://www.geekdot.com/helios-2.html

Very neat find - nothing I'd ever heard of before. I agree with Gemini000 as to why it flopped - highly parallel and proprietary machines tend to be tough sells. As far as the MFlops performance comparison, some digging on the site NJRoadfan linked brought this page:
http://www.geekdot.com/kerberos.html

I'm not sure if that 200 MFlops # should be considered "marketing flops" or "real world" for the Atari though.

I didn't realise anyone else on here had any knowledge of the Transputer! It's a bit of a favourite of mine.

In some ways it's an example of a typical British success and failure - the Transputer came from INMOS Technologies, a very early pioneer in memory technology (SRAM, for example), they had funding from the government at the time and experienced a great deal of early success but during Maggie Thatchers reign it seems they were no longer favoured as the technology darlings that they one were, so they started to suffer from under-investment in taking the design forward. The various Transputer chips didn't really advance at the speed relatively standard designs like the x86 and various RISC designs were making.

None the less, they found their way in to some fairly massive systems (thousands of processors!) - mostly confined to the sciences and engineering.

The basic premise of the Transputer was that the chip was a relatively simple design, with on-chip RAM and up to 4 high speed (for the time; 5, 10 or 20mbit) serial connections; in effect a very simple self-contained computer that just needed a few connections such as power and a 5MHz clock signal. Each of the 4 connections could link to another chip, so it was very, very simple to build various topologies of transputer 'networks'. Each chip had basic scheduling functionalities built in, and could route messages to other transputers. Even more unique was that each chip could boot the code it ran by reading a series of bytes over it's serial link - no ROM or BIOS equivalents needed.

Initially it was just Occam that they ran, but later you could write in pretty standard C, Forth and other languages. There were also a number of Unix-like operating systems written specifically for the Transputer; Helios, Parix and versions of Minix, plus the version of Unix that came on the ATW.

Anyway, enough of all that... the real reason I cast 'Thread Necromancy - Advanced' on this thread was that I actually came into possession of a Transputer interface board for the PC a couple of years ago. Almost without exception all the boards are ISA, so it relates to Vogons as it currently resides in one of my retro computers (an MB800H 845G, 3 PCI, 3 ISA, 1 AGP). It has an on-board Transputer, a 32bit, 25MHz T805, along with 16x 30 pin SIMM sockets and space for 4 extra Transputers to be plugged in; I've currently got 3 extra Transputer cards (called TRAM's in Transputer speak) installed, each with a 32bit T800 class processor and 4MB of local RAM on each. The unique thing about most of these interface cards is that they can all interface with the standard Transputer links to other, standalone systems (such as those 1000-processor examples I mentioned earlier).

I'm one of a few people still with kit in working condition; I've also decided to update an ancient driver for this hardware on Linux (https://github.com/megatron-uk/INMOS-Link-Driver), as well as starting to rewrite a few of the core tools in Python (so that we're no longer tied to compiled binaries), so it's actually now possible to use this 20-30 year old kit on a 2016 release of Linux. Of course the original 1980's DOS utilities still work, but hey, the world has now moved on from DOS 😉

Sadly, almost all Transputer stuff has virtually dried up now - it's incredibly difficult to find and a few people bought pretty much all the stock of the few companies in the late 90's, early 2000's, so the only kit that now hits Ebay is extortionately priced.

Were there rad hard versions designed for space applications as well?

I believe so, and Wikipedia shows at least one platform (https://en.wikipedia.org/wiki/Myriade) that is still using it. I'd bet there's quite a few still in industrial applications, as that tends to be the type of stuff that shows up on Ebay, attached to various VME boards, IO controllers and the like.

Megatron-UK,

Checked out your site, lots of interesting stuff.

I'd love to be able to design transputer boards to use with my Tandy 1000tx but wouldn't have the slightest clue how to begin. And I'm stuck with an 8 bit bus. 🙁

These old-school INMOS transputers sound very similar to the stuff that Adapteva is doing these days with their Parallela platform. (check out http://www.adapteva.com)

As they say, history doesn't repeat but it does tend to rhyme.

I think that as time goes on we will start to see systems with many hundreds or thousands of cores in a mesh network. Whether it is Intel making them or someone else, who knows.