Pilot ACE

Pilot ACE

Pilot ACE — the pilot model of Alan Turing's Automatic Computing Engine — is the machine that Turing designed and bureaucracy nearly prevented. Turing produced the ACE specification at the National Physical Laboratory in 1945–46, a design detailed enough to be a working blueprint; NPL's delays so frustrated him that he left for Manchester in 1948 before construction finished. The smaller Pilot Model that eventually emerged, built under James H. Wilkinson with Donald Davies, Harry Huskey, and Mike Woodger, ran its first program on 10 May 1950 — making it one of Britain's first working stored-program computers, and the only one directly descended from Turing's hand.^[1]

Turing's original ACE concept was, characteristically, ahead of the hardware around it — his 1946 report specified a subroutine discipline built on BURY and UNBURY: a return-address stack kept in memory with its pointer in register TS31, a design that anticipated structured subroutine linkage by years. The Pilot Model was the pragmatic cut-down that could actually be built, and even so it became the fastest of the early British machines: 1 MHz, approximately 800 vacuum tubes.^[2]

Arithmetic by software

The Pilot ACE shipped with a memorable hole: no hardware multiply or divide. Multiplication started out as fixed-point software — which immediately proved unusable because numbers escaped range — so Wilkinson's team wrote floating-point routines instead. The experience made Wilkinson the era's leading expert on floating-point error analysis and produced his classic book on rounding errors — a career's worth of numerical-methods insight that started with a missing instruction.^[3]

The hardware that did exist was finicky in the instructive way of delay-line machines: instruction timing depended on where the instruction physically sat in the mercury lines — an addition could take anywhere from 64 to 1024 microseconds depending on when the word came around. Main memory was 128 32-bit words, later 352, with a 4,096-word drum grafted on in 1954 — which is when Brian Munday's General Interpretive Programme arrived, a matrix-package system whose "bricks" read their own array bounds off the drum and could therefore run on any size of matrix without recompilation. Woodger's "block floating" storage scheme — one shared exponent per array — was invented for it.^[4]

From prototype to product line

Though intended as a proof-of-concept, Pilot ACE was useful enough to go into service in late 1951 and work for four years — and useful enough that English Electric turned it into a product: the DEUCE, the commercial version sold from 1955. The Pilot machine itself was switched off in May 1955 and given to the Science Museum, where it remains — one of the few first-generation British machines you can still stand in front of.^[5]

Where it sits

Pilot ACE is the "what might have been" machine: a Turing original that arrived late, scaled down, and still managed to be Britain's fastest early computer, a commercial ancestor (via DEUCE), and the training ground where Wilkinson became the world's expert on floating-point error. Its designer never saw it run — the machine is the artifact, and it outlived him.

Check your understanding

• Who designed the ACE, and who actually built the Pilot Model after the designer left?
• What two arithmetic operations did Pilot ACE lack in hardware — and what did that absence produce?
• Why was instruction timing variable on the Pilot ACE?
• What commercial machine descended from Pilot ACE?