Today, SoCs (System on a Chip) have become common in all kinds of devices, from smartphones to televisions, video game consoles or computers. The industry seems to be very clear that the future is to have the more the better within the same chip. But it was not always like this.
Our story begins in the early the 80’s. Personal computing was taking its first steps and engineers ran into a major problem: the technology of the time was impossible to increase the number of transistors of a chip with the necessary speed and CISC designs were stagnant, so many went for parallel processing: divide the work among multiple processors was the most feasible option and everyone was convinced that the software of the future would be prepared for it.
Almost by surprise, Inmos, a small British company based in Bristol, surprised the world with the first general-purpose microprocessor designed to be used for parallel processing: the transputer. David May Y Robert Milne They were the ones who made the miracle possible.
The name transputer combined the concepts of transistor and computermaking its purpose clear: in the same way that transistors form a processor, transputers would be the basic unit, the “blocks” with which build very powerful computers at low cost. Parallel computing would do the miracle.
Inmos’ objective was to produce a catalog of transputers with different characteristics and prices that included, on the same plate silicon, one CPU, memory RAMsupport for buscommunications Serie and even his own operating system in real time (RTOS). To make them work, it was only necessary to provide them with electrical power and a clock signal for coordination of operations.
Thus, at a cost of only a few dollars per unit, the transputers could be used as the main CPU (or part of it), but also as a disk controller or to control communications. While in traditional systems the processor dedicated to the disk controller, for example, sits idle when not accessed in a transputer system, those processor cycles can be spent on other tasks, dramatically increasing the overall performance of the machine.
Already in its basic design each transputer could connect with four otherswith bandwidths of up to 20Mbit/s (very high for those years). Each group, in turn, could connect with others, in links of up to tens of meters to build powerful computing farms.
Transputers connected in a fixed point-to-point pattern, so communication between two distant ones required going through all the intermediates. This caused delays in large networks that Inmos tried to solve by integrating a zero delay switch capable of interconnecting up to 32 transputers at the same time.
The “transputers” were designed to be programmed with the Occam language, developed by David May and based on the calculation of sequential processes. Inspired by CAR Hoare’s CSP (Communicating Sequential Process) communication paradigm, it supported concurrency and communication between processors. Subsequently, implementations of more conventional programming languages were released, such as C, FORTRAN or Pascal.
Inmos announced the launch of the first transputers in 1983, though they weren’t available until the following year (for reference, the original Apple Macintosh was also released that year). They included 2KB of RAM and an integrated memory controller, so more could be added without the need for additional hardware. The 16-bit T2, 32-bit T4, and 32-bit T8 series with 64-bit floating point support were released.
transputers lacked support for virtual memory, which prevents the adaptation of the variants of Unix operating systems that began to become popular (there were adaptations of variants, such as Minix or Idris). Commercially, Inmos stayed in no man’s land: Although the performance was very good for workstations when it was released, there were already RISC alternatives that surpassed it. In personal computers, dominated by 8-bit and 16-bit machines in a fierce competition arena, any Inmos option was too powerful and expensive for users.
Even the supposed advantage of building machines by combining blocks vanished when the T800 hit the market for $4,400 per piece, making any alternative price/power comparison unfeasible. The Atari Transputer Workstation, released in 1989 and based on the T800, is the classic commercial transputer-based example.
Inmos had time to make one more attempt: the T9000 had many features in common with the T800 with the addition of supporting superscalarity. It featured a real high-speed 16KB cache instead of RAM and major bus improvements. However, the funds ran out, sales of the previous models did not go well, and finally, the company was acquired by SGS that eliminated the T9000 and opted for less innovative projects, but more profitable as well as the construction of microcontrollers for consumer electronics.
For a few years in the late 1980s, the transputer was considered to be the architecture that would mark the future of computing. Neither Inmos nor the transputer succeeded, but we must recognize that influenced the development of new ideas, several of which are part of the systems we all know.
More information:
https://www.atarimagazines.com/startv4n4/transputer.html
http://people.cs.bris.ac.uk/~dave/index.html
https://www.transputer.net/tn/06/tn06.html#x1-20001
