If you think the words “supercomputer” and “punchcards” don’t belong in the same sentence, you haven’t visited Seattle’s Living Computers Museum + Labs lately. One of the world’s first supercomputers has been resuscitated there and is capable of doing useful work again.
RELATED STORY: Inside Microsoft co-founder Paul Allen’s newly expanded Living Computers Museum + Labs
Well, so long as your work isn’t urgent and requires only 500KB of memory and 3,000 times less processing power than your smartphone. Don’t laugh. The 10,000-pound, room-filling Control Data Corporation 6500 was ten times more powerful than anything else in its day.
The only one of its kind in working condition, the CDC 6500 is running — intermittently — at the newly expanded and renamed museum, and the public can sign up for remote access time on it, just like in the good old days of timesharing.
“Restoring this thing has been my life for the last two-and-three-quarters years,” said Bruce Sherry, the museum’s principal engineer.
In this GeekWire video, Lath Carlson, executive director of of the museum, takes us up close with the restored supercomputer.
The computer was built in 1967 by supercomputing pioneer Seymour Cray and was retired in 1989 after 20 years at Purdue University. Museum founder Paul Allen bought it for an undisclosed sum from the Museum of Industry and Technology in Chippewa Falls, Wisc.
During its first lifetime, the machine aided research into the cause of the common cold and through time-sharing helped in nuclear-physics research, cryptography and other government-related projects, said Lath Carlson, the museum’s executive director, during a press tour this week before the museum’s reopening Friday.
When it arrived at the museum in 2013, it was in good condition, with no broken parts or corrosion, but a number of cables had been cut and the refrigeration system needed restoring. Power it up without cooling and “you’ve got a mess on your hands,” said Robert Michaels, who was hen the museum’s engineering manager but has now retired. The original refrigeration has been restored, and the room itself is air-conditioned, too.
The cables had been cut because “no one thought it would ever be used again, and they wanted to take it apart and get rid of it as quickly as possible,” Sherry said. “Taking it apart would have taken two weeks otherwise.” Restoring all the cables “was quite painful. There had been three makers of the cables, and all of them were out of business. So I had to buy 50,000 connecting pins at 75 cents each, even though I only needed 10,000, and make my own replacement cables. It took one person five weeks to rip out all the old cables and put in the new ones.”
Processing power is provided by about 5,000 modules, each with 64 individual transistors. The transistors are configured in two boards facing each other, a construction referred to as “cordwood.” This construction packs the transistors in so tightly that museum engineers had to design special tools to reach in and replace them when necessary. The module fronts are aluminum, which conducts the cold from the circulating Freon. The museum has reverse-engineered the modules to produce new ones when necessary.
Memory is provided by 5-inch magnetic-core memory squares, each made up of 4,000 tiny donut-shaped pieces of iron. The squares are hand-woven on two axes with thin copper wire. Changes in the current passed through each donut gives it the value of either 1 or 0. Women laboring under microscopes wove the screens by hand. They’re astounding to see, beautiful in their way, and yet extremely unreliable, Sherry said. “These are the most problematic parts of the machine. Eventually we’ll build replacements, and the machine will be reliable.”
The modules and the RAM are configured into two processors, analogous to two cores in a contemporary CPU chip.
Programs are run by processing stacks of 80-column punchcards. “Everyone from that era has a horror story about dropping or losing a stack of cards with their Ph.D. thesis or research project on them,” Carlson said. “You’d bring the stack to the computing room and leave them outside the room to be run. After they were run, you’d get back either a printout of the results or a note that you had an error. If you had an error, you had to go through all the cards, find it and resubmit them.”
Much of the noise and vibration in the museum’s huge computer room seemed traceable an adjacent card-sorting machine, which like the CDC 6500 periodically stops functioning. When it does run, it’s probably the fastest thing in the room, yet all the speed serves only to mechanically propel punchcards past a light that shines through the holes to input their minuscule data loads.
The CDC 6500 is running an operating system called NOS (Network Operating System), a command-line interface (obviously). A video console does provide a graphical interface of sorts, but it’s available only to the machine’s administrators, not to end-users, and it has to be programmed just like anything else running on the machine.
Logging onto the machine can yield useful work, depending on how that’s defined, Carlson said. “People can reproduce work from their days as a student or just try to figure out how the machine was set up and used,” he said. “It’s good for a more historically oriented inquiry.”