A long time ago I heard an anecdote about an IBM corporate computer where they ordered a memory upgrade, and the service engineer came by and simply flicked a switch in the enclosure. Apparently the memory bank was already installed but deliberately deactivated until the client paid for it.
I’ve tried to find confirmation about this anecdote, but didn’t manage to. It would have taken place before the nineties, maybe seventies? I’m not sure whether it was a mainframe or a smaller computer. And it might even be a different brand of computer.
I thought I’d throw the question out here as I know there are several posters with vast experience in the computing world during the relevant period.
Incidentally, I just checked what ChatGPT had to say: it referred to an IBM PCJr, but that was a case where you installed a memory card that you activated with a switch, which is not what I asked about. When pressed about a corporate computer, it came back with the System/38 having something called ‘memory sizing’ with an ‘unlock’. That sounds like what I meant, but I don’t trust ChatGPT and can’t verify what it said about System/38.
Considering that IBM mainframes at the time (before the 1980’s?) had core memory, which cost a fortune because each bit was woven by hand, I seriously doubt it. Core memory was tiny magnetic donuts with a grid of wires and a sense wire, and threading those fine wires was a manual job done in the far east.
There was the Amdhal computers - the first IBM clones, copies of a 370-type mainframe. Amdhal was an IBM engineer who saw an opportunity to undercut price, since IBM used mainframes as a cash cow. Amdhal made only one model of their computer (simplicity!), and if you ordered the less pricey one they inserted a card to “steal” some of the CPU cycles so the capacity available was less. Our main systems engineer complained that when they asked to upgrade the power of the computer, the Amdhal tech came and took out a board.
(They were negotiating once upon a time to upgrade the mainframe, and the IBM guys would budge on lease terms. So they threated to go to this new company, Amdhal. IBM said “you wouldn’t dare!” They did. Next time the lease was up, IBM offered them a much better price.)
IBM made huge profits off their bigger computers, typically leased with expensive maintenace support agreements. The story goes when the first IBM AT computers with 286 chips at 10MHz were available, the System 36 guys had a coniption. “You can’t sell a ccomputer for $7,000 that’s as fast as our $100,000 System 36 minicomputer!!”
There was an old IBM line printer (1403) that could be sped up with a field installable upgrade. The technician simply replaced one drive chain for another.
It is true that the IBM stuff is not free software/hardware, so even today you have to license ($$$$$$) however many processors/cores and specific software products that you want to run on them. It is basically analogous to “flipping a switch” if you want to put it that way; more like agreeing to their terms and purchasing a valid license code, without which your workload will not run, at least not without violating the terms you agreed to.
IIRC the old XT clones of an IBM PC required dip switch changes.
Simms weren’t available yet. I remember plugging in IC chips to add memory from 256k to 512k. It still wasn’t recognized until setting the dip switches.
The IC chips were 32k or 64k? There were 4 or 8 in a long plastic tube. (256k total) Needed a ground strap on my wrist before removing and handling the chips.
Set the dip switches with a ballpoint pin.
I think the 286 pc’s had dip switches for memory too. It was the 386 or maybe 486? that recognized and tested memory in POST.
IIRC the line printer had a chain with characters on each link. There were 132 hammers, one for each character position on a line of that wide fanfold paper with holes up the sides. When the correct character was in front of a hammer, it would punch that character against an ink ribbon onto the paper. When the whole line had been printed, the paper advanced a line. I presume changing the chain meant a new one eliminating less used characters and distributing more of the more common letters along the chain, so lines were printed faster. (Odds the needed character was under the hammer happened sooner). I also recall switching the chain when it was necessary to print lower case, which went slower with more different characters on the chain.
(Slow is relative. These things could print a page or more (80 lines) in a second, and at month-end we’d go through a few boxes of multipart carbon paper. If you had the cover open, you needed ear protection.)
To add to the OP -
Memory - core or chip - was very expensive for a long time, so pre-loading a bigger computer with extra RAM on the chance it would be asked for later was not worthwhile. As mentioned, the computers had either sockets or slots for extra RAM, and it was bought and added as needed. Also, once RAM was solid-state chips, the capacity kept increasing by the year, so the lifetime of a size of chip meant it would be significantly cheaper a year or two down the raod.
I was in intern at IBM back in 2001. The product I worked on was a server where, depending on the configuration, you either could have 32G of RAM, or 16G of RAM in two redundant banks for higher reliability. In a sense I guess you could say you could double the RAM in that machine by “flipping a switch” at the expense of some reliability (actually as I recall it was a setting in the BIOS, not a literal switch).
(Incidentally I remember it was 2001 because that’s where I was on 9/11 – working in the lab at IBM)
One of the computers I worked on while in the navy was the AN/UYK-3 and it had core memory. One glorious day one failed, was swapped out with the one spare that we had on hand, and after making sure it wasn’t wanted back the failed one was opened. It was an 8-inch cube something like this.
For a while, Intel sold processors with half the cache memory disabled. If you paid them $30, they’d send a software patch to unlock the extra cache.
Tesla has sold cars with part of the battery capacity artificially limited. If you paid a fee, they’d unlock the extra.
Intentionally disabling fully functional resources (memory or otherwise) is incredibly common in the electronics industry and elsewhere. But it’s much less common for companies to unlock the extra for a fee. Probably because they don’t want to remind people that the hardware was already there.
Back when working for a competitor I heard this about an IBM card reader; supposedly there was a switch to double its speed. How true that was I never found out.
Lots of IBM mainframes were leased from IBM or intermediary companies. The lessees never touched anything inside the racks, only authorized service personnel would do that, so there’d be no worry about unauthorized enabling of additional installed memory, if it was done. But memory was expensive back then, even after core was replaced by chips so it seems unlikely it would remain installed in a machine producing no money for the owners.
Perhaps a switch to cut the speed in half to prevent cards from jamming up the reader. We duped a lot of decks as they were read so a clean deck could be read the next time.
Leasing is one of the key points about many computers back in the day. A leased machine could reasonably be delivered with a specification higher than the customer specified. Especially if there was a good chance that the customer would pay for added performance when needed. It wasn’t unheard of for customers to need extra capacity for specific short times (like end of financial year). Customers could be computer services bureaus rather than just a single company running its own payroll or inventory. Keeping the machine busy and making money was critical, and an ability to scale just good business sense. Making computers that could scale similarly good business. Especially with the margins on computers back then.
Computer companies became leasing companies, and in some cases found they made more money leasing than making computers. Towards the end, Control Data announced they were restructuring to be a primarily finance focussed company. They went under not too long after. They sold off the last of their money making hardware divisions - disks - to Seagate in the late 80’s. IBM was making more money on services than machines for a lot of its time. (Probably still do.) So much so that at one point they were Sun Microsystems biggest customer. They really didn’t care whose hardware the contract demanded, the meat of the profit was the services.
At the big end of town the hardware was only a small part of the deal, and the salesdroids were the ones that called the shots, not the engineers. *
Core memory is interesting. Costs per core dropped dramatically over the technology’s lifetime. Towards the end it was made with automated machinery and with remarkable density. Tricks like shooting tiny needles with compressed air through the cores helped. I have a few core memory units. One from the early 60’s and very clearly hand made, with big chunky cores - much the same, maybe a bit more recent, than the one DesertDog links to above. Another from the mid-70’s with cores so small you really need a magnifying glass to make them out. Quite remarkable. There were efforts to make core memory with thick film integrated circuit technology. I have a board of memory from an ICL machine made this way. But all a little too late for core.
From the mid '70s onwards dynamic RAM started to beat out other memory. Core held on in some niche areas - where its ability to maintain data without power was important. It was generally very resilient. The AN/UYK-3 linked above was one in a line of naval computers, cumulating in the AN/UYK-43 - which still used core memory and saw shipboard service right through the 80’s. For use cases like the NTDS the navy really wanted maximum resiliency.
* Favourite joke:
An IBM salesdroid and an IBM engineer go on a weekend trip bear hunting.
They arrive that evening at a small log cabin in a clearing, deep in the woods.
The next day they get up bright and early, and the IBM salesdroid heads off for a walk. About an hour later he comes running and yelling at full pelt out of the woods, back across the clearing and straight in the front door of the cabin - with a huge, very angry grizzly bear a few paces behind. He slams the front door, runs around the cabin table with bear right behind, then out the back door, yelling to the engineer as he slams the door: “You sort this one out, I’ll go rustle up another one.”
The worst example I know about is BMW trying to have a subscription for heated seats. They dropped the idea, thankfully, but damn. I don’t have a particular problem with most software-locked features, but that one crosses a line.
First solid-state memory was in the System/370 model 145 in 1970. When I started in 1980, we had a System/360 model 75 with a whopping 3MB of memory: one MB core, two solid-state. The big machine! Core was mostly made in Mexican maquiladoras. And some by Navajo. Always women, apparently—it was essentially weaving, and women had the skills for it.
I never used an Amdahl (note correct spelling) but had heard that there was some version that was upgradable by cutting a wire. Still looking for a cite.
Make that “always leased”, until the 80s. And yes, the competition between the divisions was brutal. There were stories of customers getting visits from IBM sales droids from each division, who would assert that the other reps were selling junk. IBM figured this out around 2000, reorganized sales.
One old-timer recounts:
“The 370/165 had 2 microsecond core memory (4-way interleaved), but various competitors came along with solid state replacements, which could often run somewhat faster. Turned out the 165 had a small board that contained countdown logic to wait for the memory to be ready, and by simply unplugging that board the countdown would be eliminated (or rather, reduced to the number of ticks needed for the not-yet-released /168 with solid state storage), so removing the board for normal use and plugging it back in before calling IBM for service became just a part of normal operation.”
Also relevant to this thread–look for “accelerator”:
