Article 01 · Oct 2026 · Machine story

The Cube That Wore Its Wiring on the Outside

A black cube with a red heartbeat

Some old computers are beige boxes you forget the second you walk past them. The Thinking Machines CM-2 is not one of those.

Picture a black cube about five feet tall, built out of eight smaller cubes. Behind its dark panels, a field of tiny red lights blinks and glows. It looks less like office equipment and more like something that landed in the lab overnight and is quietly thinking about you.

That is pretty much what its designers were going for.

When I shared this machine here a few weeks ago, a lot of you lit up about it (yes, that pun was intentional). So I wanted to give the CM-2 a proper sit-down, because the story behind those lights is even better than the lights.

Optional: drop in one personal line here if you want, like the first time you saw a photo of it, or how it felt compared to the machines you grew up with. Or delete this and leave it out.

65,536 tiny brains working at once

Thinking Machines Corporation was founded in 1983, growing out of Danny Hillis's work at the MIT AI Lab. Hillis wanted a computer built for the kinds of problems a human brain handles easily and ordinary computers struggled with, like recognizing faces.

Most computers of that era did one thing at a time, really, really fast. Hillis went the other way. The CM-1 arrived in 1986, and the CM-2 followed in 1987 in the same famous cube, now with optional floating point hardware for heavy number crunching.

A fully loaded CM-2 had 65,536 processors. Each one was almost comically simple. It worked on data one bit at a time and had its own 8 KB of memory. On its own, one of these little processors could not do much. Together, all 65,536 could perform the same operation at the same moment, each on its own piece of data.

Think of it less like one genius and more like a stadium full of people all flipping cards at once. That is the "data parallel" idea, and it was a great fit for things like fluid flow, simulations, image work, and searching through mountains of text.

The connections were the point

Here is the part I love. The name "Connection Machine" was not marketing fluff. Hillis put the emphasis on how the processors talked to each other, not just on the processors themselves.

The processors were grouped sixteen to a chip, for 4,096 chips in a full machine. Those chips were wired together as a 12-dimensional hypercube. That sounds like science fiction, but it just means each chip was wired directly to 12 others, so any two chips could reach each other in 12 steps or fewer.

Twelve dimensions is a lot to wrap your head around. Even the people building it thought so.

Tamiko Thiel, who led the packaging and industrial design team, has written about being overwhelmed trying to picture all that wiring. So she asked Richard Feynman, the Nobel Prize-winning physicist who was helping with the network. (His son Carl worked on the Connection Machine, and Feynman got so curious he asked to help too.)

Feynman's answer was basically "Oh, that's easy." He drew a line, then a square, then a cube, then a cube inside a cube. And then he said that after that, it gets too hard to draw.

Thiel kept going. She found a way to draw the higher dimensions as repeating cubes of cubes. A 6-D hypercube looks like a cube of cubes. A 12-D one is a cube of 9-D ones. That drawing became an internal logo, then a T-shirt, and eventually the shape of the actual machine.

So when you look at a CM-2, the outside is literally a picture of the inside. The cube wore its wiring diagram as a jacket.

Ok, so what were the lights actually doing?

This is where it gets fun, and where it is easy to get things a little wrong.

There was not one light per processor. There was one red light for every 16 processors, so one per chip. On a full machine, that works out to 4,096 lights.

Thinking Machines' own 1987 technical summary spells it out. The red lights on the cabinet could display internal status information, or they could be handed over to the user's program. A command called set-system-leds-mode picked which, and another called latch-leds let a program switch the lights on and off directly.

So the honest answer is: both. By default, the lights reflected what was happening inside the machine. But programmers could also take the wheel and make the lights do whatever they wanted.

And people did. There is a well-known story, repeated in Wikipedia's write-up of the machine, that visitors wanted to see the lights blink, and so plenty of finished programs ended up with extra operations just to make the LEDs put on a show. I cannot point you to a primary document proving how common that was, so take it as lore. But it is very believable lore.

One small myth-buster while we are here. You may have seen that hypnotic, wandering "random and pleasing" light pattern. That belongs to the CM-5, the later Thinking Machines supercomputer with big flat LED panels (the one in the Jurassic Park control room). One of the people who says he designed those CM-5 panels described a rotary switch with a "random and pleasing" mode. The CM-2's lights are a different story, tied to its chips and its programs.

Hard cube, soft light

Thiel has written beautifully about what the team wanted the CM-2 to say. They did not want another beige box. They wanted the form to explain the function, even though the function was invisible.

Carl Feynman had a fantasy of the machine as a vast cloud of lights flickering as messages flew back and forth, like neurons firing. Status lights were already common on circuit boards. So the team asked, why not move them to the outside and let people see the machine work?

The result was a deliberate contrast. A hard black geometric cube. Filled with a soft, constantly changing cloud of red light. Thiel described it as a way to ornament without decorating, and I think that is just lovely.

The industrial designers Allen Hawthorne and Gordon Bruce, who had years of experience designing IBM products, helped with the detailed design. Thiel also mentions the sculptor Arnaldo Pomodoro as an inspiration, with his polished shapes that seem to hide whole worlds inside.

A supercomputer you plugged into another computer

One more charming detail. You did not sit down at a CM-2 like a desktop. It hung off a "front end" computer, either a Symbolics 3600 Lisp machine or a DEC VAX. You wrote your program on the front end, and it sent instructions to the cube, which acted almost like a giant, very smart chunk of memory.

The 1987 technical summary also lists the parallel unit at 56 by 56 by 62 inches with 28 kilowatts of power dissipation. That is a lot of heat for a bunch of red lights and some very busy one-bit friends.

Where you can see one

The CM-2 aged into a design icon. Thiel writes that a CM-2 was acquired by the Museum of Modern Art in New York and shown in its 2017 to 2018 "Thinking Machines" exhibition. Others have been preserved in museums too, including a CM-2 and DataVault at the Mimms Museum of Technology and Art in Georgia.

Every time I look at photos of it, I get the same feeling. Somebody looked at a box of chips and said, "No, this should look like what it is." And then they actually did it.

Your turn

What did your first encounter with a "serious" computer look like? Was it a blinking rack in a school lab, a terminal in a back room, or a machine you only ever saw in a magazine? Tell me what it looked like and what you thought it was doing in there.