Transputer Project

High speed data logger for PE Tech in 1992

Transputer PCB top and bottom

The Transputer project was a custom design for the Port Elizabeth Technikon for a high-speed data capture system. Due to the speed and memory requirements, the 68000 was not going to manage, so an alternative was required. We had done a fair amount of evaluation with a Transputer on a PC plug-in card, Occam and the Inmos C compiler for the Transputer. The initial work was on a T414 without floating point, but other applications like surface inversion for metal press moulds would require floating point. The top and bottom layer plots were scanned in many years later then stitched together, which is when it became apparent that the output plot routine was not very good. The specification for the double width extended VME card was:

The testing for the board involved reading in four channels at the fastest rate within a for loop to fill up the 32 MBytes of memory. A scope on the chip select showed the peak sampling rate was 3 MHz. It was a 32-bit read of 4×8-bit Flash ADCs, followed a 32-bit write into memory and incrementing an array pointer. The other 32 channels were ±10V, 12-bit resolution ADCs with simultaneous sampling by synchronising on a single chip select, then sequentially reading out of their FIFOs.

Link Adapter Interface

B004 link adapter

The Inmos B004 compatible link adapter reverse engineered on a non-critical wire wrap board. A close-up would reveal a touch of rust as it was discarded before making a proper board and Port Elizabeth is on the coast.

The link adapter board had to be reverse engineered, as there were no application notes and the Inmos representative would part with little knowledge. The link adapter ran from a 5MHz crystal and I recall it clocked out serial data at 20 MHz. Anyway, other than decoding guess work and checking for inversion, not that hard. The prices for link adapter boards were ludicrous.

Transputer’s Demise

The photo was of a blank board many years later. Four boards were made, three populated and although the prototype was sold, the other boards were given to the researchers as they could not get funding—affecting one researcher’s M.Sc.Eng thesis. We also had to buy a lot of T805s as long lead-times and minimum order quantities were standard practice. Each T805 cost the same as a washing machine, so hopefully my long suffering wife never reads this. Two years later we tried to sell them to another group near Pretoria, but they used a different package, and we never did see the T805s again. As an aside, the Inmos representative who visited Stellenbosch University around 1987/88 and refused to say how the link adapter worked, or where the stack resided after power-up, or how to boot from standard Flash at a Transputer course, hopefully he found out that to sell the device, engineers needed to know how to use the device without riduculous prices for evaluation boards that could only link to other Transputers. Inmos is no more (surprise), and delivered the T9000 too late to save their sorry backsides. Indeed a bitter lesson which made me wary of UK electronic companies for many years (ARM was another company from the people who overcharged Napolean for gumboots but they managed to pull through after the Archimedes workstation debacle and prices for C compilers.)

There were a large number of design teams in South Africa who went down the Transputer path. Academic projects received generous funding for multiprocessor crates, plus sanctions did not appear to affect trade. Momentum was quickly lost as Inmos floundered. The concept was good and performance was similar to a decent PC at the time, but obviously volumes were not enough to keep Inmos afloat.

Transputer PCB top