On November 16, 1953, a small computer at the University of Manchester ran its first program. It used 92 transistors, 550 diodes and a few vacuum tubes.
That last ingredient sounds like a disqualification. The machine is widely described as the first transistor computer, yet valves still generated its clock pulses and helped read and write its magnetic drum. The new technology worked, but it could not quite leave the old technology behind.
The future was less reliable than the past
The Manchester team used point-contact transistors: tiny germanium crystals touched by two fine wires, much like the cat's-whisker detectors in early radios. They promised smaller machines and lower power consumption. The transistor prototype used about 150 watts, while the nearby valve-based Ferranti Mark I consumed 27 kilowatts and occupied a large room.
The new components also had an awkward defect. They were less reliable than the valves they were supposed to replace.
Manchester built the machine anyway because the direction was obvious even when the parts were not ready. The Computer History Museum records a larger version completed in 1955, and the University's own history says its early transistors managed an average error-free run of only 1.5 hours. The commercial version, built by Metropolitan-Vickers, switched to more reliable junction transistors. Six of those machines went to work inside the company.
A computer that waited for its memory
The machine's other compromise was its memory. It had no fast random-access store, only a magnetic drum recycled from the Manchester Mark 1. Programs could spend much of their time waiting for the right word to rotate under a read head.
The designers made the waiting part of the instruction set. Each instruction included the address of the instruction that should run next, so a programmer could arrange code around the drum and ask for data when it was about to arrive. Even the accumulator and current instruction lived on special drum tracks, continually copied forward as the cylinder turned. The University's technical account calls this "optimum programming," which was less an optimization pass than a willingness to think like rotating metal.
The first transistor computer was therefore a hybrid in almost every useful sense: new switches with old valves, electronic logic with mechanical timing, and a promising future assembled from unreliable parts. It did not prove that transistors had already won. It proved that they were worth building around before they had.