In parallel to the Meg project it was decided to build a relatively small and economic computer. When design was started in 1952 it was clear that the project could provide valuable experience in the use of the recently introduced transistors. It was built even though the germanium point transistors were more unreliable than valves, because semi-conductors held out the promise of lower power consumption, higher operating speeds, smaller size and greater reliability in the future.
The experimental Transistor Computer was first operational in November 1953 and it is believed to be the first transistor computer to come into operation anywhere in the world.
There were 2 versions of the Transistor Computer, the prototype, operational in 1953, and the full-size version, commissioned in April 1955.
The only storage used was a drum (rescued from the Manchester Mark 1 !). This meant that the average random access time to a word in store was half a drum revolution, i.e., with 64 words on a track, 32 times the random access time for a word if it could be stored in a true RAM. So the Transistor Computer was slower than the Mark 1. Both versions had a pseudo 2-address (or 1+1) instruction format, where the address of the next instruction to be obeyed was contained within each instruction, to facilitate "optimum programming". The drum was even used to store the Accumulator and the Current Instruction. Each used a track of its own on the drum, and a special "regenerative" system to ensure that access time was fast.
The word length was 44-bits, divided into 4 "syllables" for an instruction.
The prototype computer (Nov. 1953) had a simple 7 function order code and one track of 64 words for main storage. For the full-size computer (Apr 1955) the order code and storage were much extended and a hardware multiplier included. A third "regenerative" drum track formed an 8-word B store. This provided a set of registers with faster access times, which could be used both for instruction modification and any arithmetic or logical operation except multiplication. Arithmetic was serial, with a pulse rate of 125,000 per second. The instruction times were directly related to the 30 millisecond drum revolution time (the basic unit being the time to read a word, i.e. 1/64th of a revolution).
The 1955 machine had a total of 200 point contact transistors and 1300 point diodes and had a power consumption of 150 watts. There were considerable reliabilty problems with the early batches of transistors and the average error free run in 1955 was only 1.5 hours. The design of the full-size Transistor Computer was subsequently adopted by the Manchester firm of Metropolitan Vickers, who changed all the circuits to more reliable types of junction transistor. The production version was known as the MV950 and was completed in 1956. Six machines were made, all for internal use within the company.
The most important impact of the Transistor Computer was it provided early experience in transistor circuit techniques.
The faster access required for the special registers, i.e. the accumulator,
the Current Instruction and the set of 8 B-registers, was achieved by special
"regenerative tracks" on the drum. Each track had a "write" head placed a
suitable distance back along the circumference from a "read" head, e.g.
(on the full-size machine) one
word length for the Current Instruction track, two for the double-length
accumulator track and eight word lengths for the B-track.
Then the information on each track was continually regenerated by reading each
bit and copying it to its next position back down the track, unless it belonged
to a word that was being rewritten. So at the end of each basic unit of time
(1/64th of a revolution), there was no wait time to access the
Current Instruction, at most one unit of time wait for the double length
accumulator (or a particular half of it), and at most 7 units of time wait
for a particular B-register.