EDSAC remembered with tubes of liquid mercury; the museum rebuilding it swapped in steel wire

EDSAC kept its numbers moving through tubes of liquid mercury, and its rebuilt twin keeps them moving through wire.

The National Museum of Computing calls EDSAC “the first practical general purpose stored program electronic computer,” a title it admits is the “generally accepted” one. Built at the Cambridge University Mathematical Laboratory, it ran its first program on May 6, 1949 and then kept a service going for over nine years. The museum describes the memory as “delay line tanks,” which it defines as “metal tubes filled with mercury.” They worked because sound crawls through liquid metal. Slow is the feature here: a pulse that takes its time crossing a tube is a pulse that is still in the tube.

Mary Bennett’s write-up for Embecosm lays out the loop. A binary word goes in as pulses, travels the mercury, gets received at the far end, is turned back into binary, and is sent into the start of the tube again. Nothing sits still. The data lives in the trip, like a conga line that has to keep dancing or it forgets who’s in it. She puts the original at “over 30 mercury delay lines all stored in a coffin-like box,” and says most tubes held a multiple of 36 bits, a long word. How many bits a tube held depended on its length and the pulse frequency.

The machine those tubes served was small by any measure. The museum lists 512 words of memory at first, later 1024, only 18 operation codes, and about 650 instructions per second. That’s the whole spec sheet for a machine Maurice Wilkes designed in 1947 (your thermostat has more).

Rebuilding it brought a plumbing problem: mercury. The museum’s own post on the subject, “Delay line storage in EDSAC - pick your poison,” says copying the tanks was “not practical to replicate this in the reconstruction because of cost, safety and reliability.” The dad joke was already in the title, so I’ll leave it there. Bennett came to the same place on her own project and wanted to avoid “mercury poison” and the “large bill” that came with it.

The replacement is spelled out on the EDSAC page: “Our alternative is to use magnetostrictive steel wire delay lines.” The museum says the wire lines work on similar physical principles and connect to circuits identical to those used with the mercury lines. The 2020 post adds that a volunteer, Peter Linington, was building nickel delay lines with steel wire as the medium, to replace temporary semiconductor stand-ins used during commissioning. It also mentions the story that Alan Turing proposed gin as a delay line medium, which the museum calls “a little hazy.” Fitting, for gin.

Bennett’s own detour was air. For an FPGA version of EDSAC built for the Chip Hack EDSAC Challenge, she made a delay line out of a tube 1 meter long and 11 cm across, standing in for the machine’s 72-bit accumulator. Air costs nothing, nobody has to sign a hazmat form, and sound still takes its time.

The museum says the rebuild is “close to 100% physically complete” and in commissioning, with Marshall Amplification having built almost 20 memory unit chassis. The EDSAC page gives no memory capacity for the rebuilt machine.

A reconstruction that must stay running for decades will probably keep making this kind of trade: same logic, safer hardware. Because the wire lines plug into circuits identical to the mercury-era ones, the delay medium can change without the rest of the machine noticing. Other rebuilds with a toxic part on the parts list will likely end up in the same place, and the ones with a volunteer who builds the replacement will get there sooner.

Pick your poison. They picked wire.