Core Memory questions

I remember there being on a shelf in the computer lab this square of what honestly looked like a mix between finely woven chainmail and a quilting square but it was apparently a demonstration piece and or broken as there were no connectors no nothing that looked like it could plug into anything. All the descriptions i’ve seen of core memory makes it feel like a mix between a solid state drive and sleep mode in that you could unplug the machine, or memory unit, or whatever, ship it to wherever, and so long as the memory hadn’t been degaussed you could pick right back up where you left off on powering on.

I have thoughts but at the same time who here has practical experiance?

Could you, as example, have something in a core mmeory setup then move to more modern simi-conductor memory for actual working ram so that the computer would boot into a specific thing that would stay in core memory but you’d have ram for either other tasks or for anything overflowing or… .just ot have ‘in case’? IE bootloader is in core memory and it then hands off to OS or program or whatever and isn’t wiped or accessed til next startup.

1 Like

Both the PDP-8 and the PDP-11 started out with core memory, but later had soild state
memory added on. You may find a model that supports both.
PS. Dr Who did just that in Logopolis . The last Tom Baker serial, 1981

1 Like

Not from personal experience, but here’s a data point on non-volatility and a system having still a valid memory image after having been shipped to a new location:

The PDP-1 that originally went to Bolt, Beranek & Newman to be used in a more formal environment than at MIT was eventually retired to the University of Massachusetts at Amherst in 1968 during my own tenure there and was connected to a model railroad. It still had SpaceWar! as the primary program.

(J.A.N. Lee, Head of the CS department at Amherst)
So this actually happened.

I’m skeptical regarding mixed memory technology, since core memory requires quite elaborate control and amplification hardware. I guess, it’s not something you’d want to do, if you can avoid it. But I’m not an expert on this – or even any of the electronics involved in this.

Edit: There’s a popular notion of PDP-1s having been shipped by DEC with Spacewar in core memory for testing, but I think, this is really a myth. First, I can’t find any source for this (every time, I asked, it turned out to be just something someone had heard of), and you’d probably want to run some memory tests for acceptance tests first, as soon as you’ve the unit installed, at which point any memory image would have been destroyed. Moreover, since a paper tape doesn’t take that long to load and isn’t that heavy in shipping, there would be little advantage in preloading that particular program. The above statement by J.A.N. Lee may well be, where this started, but the context is quite different.

1 Like

My PDP-11/05 (c. 1973) has a full complement of working core memory–all of 56KB! This is thanks in part to a ME-11 memory expansion box I have which contains an additional 40KB beyond the base 16KB housed in the system box.

You can see my system in action running an early version of Unix here: PDP-11/05 at VCFWest 2023.

Here’s a picture of me debugging one of the memory boards. The core memory plane is the board with the black rectangles on top:

Core memory itself is indeed non-volatile. However, at least on the PDP-11, the internal state of the CPU (e.g. the program counter, general registers, etc.) is stored in volatile memory, and thus is lost when the system loses power. It is possible for the system to receive an interrupt when power is about to be lost. This makes it possible for specialized software (typically included as part of the OS) to quickly save the state of computation in a way that can be recovered when the system reboots.

I saw this in action on my 11/05 when the power company decided to bounce our power while I was running a long-term memory diagnostic test. Shortly after I started the test, everything went dark. I thought I had blown a circuit, but then the power came back moments later. I happened to be using a laptop as a terminal for the PDP-11, so it never shutdown due to its internal battery. As soon as power came back, I saw the PDP-11 come to life and print a message to the terminal saying it was resuming the memory test right where it left off!

As fun as that is to see, as far as I know there weren’t many OSes that supported this feature. Certainly none of the PDP-11 Unix versions do.

As for using a mix of core memory and semiconductor RAM, this never really made much sense (at least not for primary memory). Once semiconductor memory came on the scene (very early 1970s) it dropped in price pretty quickly–so much so that by the mid- to late-70s DEC was producing PDP-11s that only used semiconductor RAM (e.g. PDP-11/03). Although static (and later dynamic) semiconductor RAM lacked the feature of non-volatility, very few systems made use of it. And those that truly needed it could use memory systems with battery backups (DEC built some of these as well).


This is why I ask the questions i do.

There is a large gap between ‘Hey that is theoretically possible wouldn’t it be neat if-’ and what people ACTUALLY did.

I think it’s cool you have a pdp/11. One of those things i kinda want but neither have the space nor even a thin justification for obtaining.

1 Like

And some neat things get lost in history.
Four phase systems had nice computer, with custom chips from 1970 and
soild state memory, not core. They never sold chips, so you never heard about them.


Plenty of system used a mix of core and semiconductor memory. Obvious early example being the EE200 which came all core, and you could add semiconductor memory if you were rich. Started with a 16byte option that covered the “registers” (they also being low memory), and a 256 byte option for the better off. All about performance. Semiconductor RAM (static especially) was much faster than core and didn’t have rewrite cycles in the same way (DRAM does but they are a bit more hidden). Later systems that wanted non-volatile elements just used batteries.

Core memory is sort of alive and well in modern different form - FRAM. Very low power, none of the flash write limits but also a lot more expensive.


Indeed. Even the PDP-11/05 which I mentioned above uses semiconductor memory for its registers. However this memory is used very differently from main/primary memory (despite the quirk of the PDP-11/05 registers being addressable in the address space of the processor–as describe here).

Could the 11 run with mixed core and semiconductor RAM cards ? The EE200 for example could as the memory bus is not synchronous (the memory provides a ready signal to the backplane and CPU board set) so it was perfectly possible to have 4K in RAM and the rest in core or similar arrangements.

… designs for sale …

See also

and I think a few others lurking out there…

Maybe not quite the densities of old, but interesting to know people are still fiddling with it…


1 Like

It could. As with the EE 200 it seems, the PDP-11 bus (Unibus in this case) was asynchronous, and thus completely agnostic to the type and speed of memory being accessed. I regularly run my 11/05 with a base of 16KiB of core and the rest of memory supplied by the modern high-speed RAM on a Unibone card. Other more period-correct options are possible as well.


Not strictly hands -on experience but I used to teach a course that include ferrite core memory principles - the students were destined to be sent to West Drayton as system engineers for the air-space control computers built by Plessey - one student (with an eye infection) said he couldn’t understand how it worked - my unthinking response was that anyone with a half - i could see how it worked - Military humour!

The Mark 1 Bloodhound system ( the radar or fire control, I believe) had a magnetic memory that used little magnetic material rods - like bits of pencil lead -
one day an instructor dropped one of the memory planes and all the rods fell out - took hours to recover them and re-insert. RAF Newton 1963.


Now I’m puzzled: how does it work, if it’s not a ring, but a rod?
(Or were these wired into a loop?)
Also, wouldn’t a rod lose magnetic polarization over time (however slowly, but relevant to the original question)?

I honestly don’t know - I wasn’t privy to that system or its course - just the tea-room mirth at his mis-fortune.

I used to work for Honeywell 40+ years ago. Both the mainframes and minicomputers used core memories on the older models. The Honeywell 516/716 minicomputers had core memory boards which could be unplugged and installed in a new machine. The field engineers who worked on the minicomputers sometimes used to load the diagnostics up on a machine, remove the core board, and take that board to another site to run the diagnostics on a different machine. I was jealous because the mainframe core memory racks would have taken a team of movers to take to a new site.