Magnetic Core Memory

Dave looks at a magnetic ferrite core memory board from 1970
Where the data is stored in individual minute ferrite toroidal rings.
Extracted from Mailbag #648 because it wasn't easy to find on KZbin search.

[@johnclawed]

Don't write to it until you read it out!!! It's non-volatile so whatever was stored in the 1970's is still there.

[@StrangeQuark]

its from a missing apollo mission

[@ronaldbrown9638]

Aliens ?? 🤪

[@streamware7746]

@@ronaldbrown9638 no, and also all Data On There is Non Existent

[@Derpy1969]

Reading it will erase it. You have to re-write what you read to restore what is in it!

[@deathbydeviceable]

@@Derpy1969 what if it was never read? But then again it is old

[@timon0x31]

I worked for a memory company in the late 70's. We were putting 64KB on a 15"x15" board. These boards ran close to $4-5,000. The wires were attached to very thin but long needles. You threaded the needles through the cores much like sewing. It was all done by hand since there were no robots at the time which could do the job. One memory board took several hours to weave. The memory had three types of cycles. Read, which also wrote the data back. write which still did a read to clear the location. Read modify write which read the data, let the processor do something with it then write the data back.

[@balajiLILG]

5000 usd ? Holy crap !

[@fernandojung5057]

Interesting, but somehow i see a comparison with a qbit...

[@mrmeval6982]

@@balajiLILG Yes, they actually paid Americans a living wage to do so-called menial work like that. I have samples of VLSI military parts that cost the US taxpayer 4,000 in 1978 dollars. Those were hand wired with solid gold wire from the die to the external connection.

[@phoenixsmith4001]

@@mrmeval6982 Why solid gold instead of copper ? Couldn't they just coat them with something like even Dave's board seems to have on it ? And wouldn't solid gold be more fragile than copper wire since it's softer ? I'd love to see some pictures if you have them on a site too.

[@jimbarchuk]

@@phoenixsmith4001 Gold doesn't oxidise, which could vary signal integrity. Gold was tons cheaper back then. This was bleeeeding edge tech at the time, so no expense spared.

[@ZylonFPV]

Omg, Please make a video where you actually get it running and save some data to it! I would love to see that.

[@georges7259]

Dave, just rustle up some sense amplifiers and hook it up to your TRS-80 Color Computer through the expansion port, which it will then boot from.The Color Computer's clock of just under 1MHz should match nicely with your 1us prediction for the memory core access speed. Some assembly (code) required :) No worries!

[@nikgolinar4378]

Zylon FPV arduino and a bunch of flip-flops?

[@ricknoah9184]

gerge s- I thought the "color computer" was a Commodore, maybe a 6809 chip.

[@MrGoatflakes]

+Rick Noah the TRS-80 Color Computer (Coco) did have a Motorola 6509 processor, like the Commodore 64, but it wasn't from Commodore and instead it came from Tandy Radio Shack. And apart from using the same processor it wasn't otherwise related to Commodore's somewhat contemporary Commodore 64, except for running, like most home micros at the time, a version of Microsoft ROM Basic. For the Cocos was this Color Basic for the lower end and earlier models and Extended Color Basic for the higher end and later 64kb versions. The Extended Color Basic ones also had Color Basic ROMs as well and I believe you could use either, although it has been a _long_ while since I have used my old 64kb Coco II. The Coco came out some years earlier than the C64. The C64 was a somewhat more able machine, although I'm sure that people will argue the toss over this as they did at the time. But there is no arguing that the out of the box graphics modes and sound support were slightly superior in the C64, especially the native graphics modes. Strangely although the Tandy TRS-80 Color Computer line was called "TRS-80" they were actually unrelated to TRS-80 Model I and Model II, which were released even earlier and which were monochrome machines based around the more expensive Zilog Z-80 processor. Instead of having an output to connect to a TV antenna jack like the Coco did they instead had their own custom monitors and had yet another version of Basic. They also I believe came with floppy disc drives as standard and were much more expensive being aimed at business. They were never as popular as the Cocos.

[@tursilion]

Almost, the Coco used the 6809, which was a different processor from the C64's 6510 (which was in turn a slightly modified 6502, the more well known variant). The two chips were different manufacturers and not at all compatible.

[@georges7259]

I took Dave's bit bucket challenge and counted 'em. Each of the 4 sections has 100 by 64 wires, or 100 64-bit words, equivalent to 800 bytes per section, 3200 bytes total .

[@EEVblog2]

I knew someone would!

[@Yamahapsr200]

well... a word is 32bytes so....

[@ryanvoots9827]

I think you meant bits, but that's not always true. Some computers had 9 bit words, 4bit, 22 bit. And everything in between, en.wikipedia.org/wiki/Word_(computer_architecture)#Table_of_word_sizes

[@ryanvoots9827]

Esp when we're talking about computers in the era of magnetic core memory. There were no standards for any of this back then.

[@nissarin]

It's actually architecture dependant and that's pretty old system.

[@abpccpba]

My source at IBM said these were hand woven. He started his career when these were in use.

[@spacepirateivynova]

'destructive' memory, which is why it has both a sense wire which would flip the bits, and an inhibit wire which would 'hold' the memory along with the x/y addressing wires.

[@stuartgray5877]

The company I currently work for (Ball Aerospace) used to make the magnetic core memories for the space shuttle flight computers. There are still techs employed that used to wire those memory cells by hand.

[@bttrs]

I always wondered if the magnetic charge could be seen though one of those green magnetic visualizer sheets. If so, you could save a bw bitmap in those beads and actually visualize if though the the green foil thingy or do an animation or something.

[@marknesselhaus4376]

I have not tried that but doubt you would see any change in the sheet as the magnetic field is circular and contained within the core. Worth a try though.

[@JLSoftware]

Fantastic! I worked on an IBM 360/67 in college. Same deal, 512K of ferrite core memory in huge cabinets. VM to 1 megabyte for your job if you wanted. These computers cost millions and took up an entire room.

[@xjet]

Remember when memory cost $7,500,000 per gigabyte? I do :-) kzbin.info/www/bejne/larTZ52qpbZonMU

[@JLSoftware]

I know it was $950.00 to upgrade my 1984 original 128K Macintosh to 512K. And never forget that Bill Gates said nobody would ever need more than 640K of RAM. At the time, that made sense because we had started out with 16K!

[@RolfMikkelson]

I thought it was IBM that insisted upon the limit, also IBM forced Intel to create the 8088 with an 8 bit bus, since "No one will ever need a 16 bit bus"...

[@JLSoftware]

IBM may have been the reason, but it was Gates who said that.

[@JLSoftware]

Your reference says he said he thought it would be enough for 10 years.

[@Mr.Unacceptable]

Take an array of LED's and get them to respond to the bit of info. You could then see visually what it's doing. You can create an ASCII picture on each panel and have it show without printing it out. Then frame it. Another idea was to have magnet paper over top and watch the beads respond to which bit is assigned.

[@andrewpike2924]

Hi Dave That brings back some old ,days. I used to work on core store memory systems. You are quite right, reading is destructive. What would normally happen is a read would involve a write to see if the core changed state, then you would do another write to set it back to its original state. In order to change the state of the core both the X and Y lines need to be activated. This is where the inhibit line comes in, it is usually in parallel with the X line. Current is passed through the inhibit in the opposite direction in order to cancel out the X line (both X and Y need to be active to set the state of the core). Where it comes into its own is when you have a bank of say 8 or 16 of these in parallel and you need to write a bit then the inhibit would be enabled on all but the bit you wanted to write. Anyway I hope I was not telling you something you already know but it is a very interesting bit of history. Thanks for the video.

[@nerdboy3458]

ur smart

[@DRAEK05]

He probably does, but the rest of us don't. Thanks for sharing.

[@Longshot762x51]

Thank you so much for taking the time to do this video. One of the original patent-holders was my Grandfather and he passed away this last April... Im an electrician by trade and was pouring over the schematic trying to understand and you explaining it out was a wonderful experience. Thank you.

[@sin3r6y98]

Modern day Dram, which is capacitor based, is also destructive on reads. There's a buffer on each memory chip that writes the data back after it's been destroyed.

[@Sekeletu]

Why would you use this magnetic memory if you could just use the memory in those chips?

[@jon87386]

Technically core memory is write-only ;)

[@opera5714]

In the late 60's I visited COGAR near Utica, NY. They had housewives weaving these core memory. Even the manager couldn't understand how they could stand it. Said they only worked about a half day, drank coffee and chatted all the time.

[@BrekMartin]

Made a 64 bit core memory from scratch a couple of years ago. One of the coolest things is writing new data in the same operation as reading the old... and erasing it with a magnet.

[@electronash]

Brek Martin Morning, mate. ;) (or afternoon? lol)

[@bonefishjr6835]

I know this comment is from 6 years ago, but if you have any resources for where I might learn more about making one could you send those my way?

[@BrekMartin]

@@bonefishjr6835My channel. Pretty average video recording then, and it wasn’t exactly a “how to”, but it’s there.

[@herbertsusmann986]

In about 1977 I had a summer job in Silicon Valley repairing older Data General minicomputers. We had 16k x 16 bit memory boards that were 15" square all core memory. It was great as you could turn the power off on the computer, restart it and your test program would still be in there. By that time however "semiconductor RAM" had completely taken over for new systems. We still shipped out used machines with core memory, however! It actually worked really well and was very reliable.

[@KiwiHelpgeek]

That brings back memories. I remember working on an old mainframe that still used core memory in the 80's! How many people figured out that the loops easily visible at the 1:38 mark were designed to provide timing delays.

[@GeoffreyFeldmanMA]

The Apollo memory was rope memory, not core memory. It was in fact read only. You are correct, most core was did destructively read but the bits would be written back at low levels of logic on the board, not CPU level usually. There was enough local capacitance to assure the write back would complete. So - at a more macro level, this was not destructive in actual practice.

[@stargazer7644]

A driver circuit would rewrite the bit after it was read out.

[@ethanpoole3443]

Geoffrey Feldman Apollo used both read-only rope memory and read-write core memory. IIRC, there was about 9 times as much read-only rope memory (~72KB) versus read-write magnetic core memory (~8KB), but like most computers it required some actual read-write memory to store calculations and variable data and to maintain the stack as well as preserve register states during interrupt events and when switching between stored programs.

[@BlackEpyon]

The Apollo computers did have core memory for RAM, as well as rope memory that functioned as ROM (where the programs were stored).

[@rkflorida4355]

I watched these being made at the Univac factory in late 1969. They were hand threaded by a bunch of ladies with very powerful magnifier lenses. Amazing stuff. Also, you have a sample of 2.5 D memory which is destructive, but a 3 D memory was also made that did not require the additional clock cycle to rewrite the bit.

[@Mythricia1988]

I have a single Univac core memory plane. Tiny stuff. Quite amusing to compare to a Soviet era slice I have - the Soviet one has less total bits, but it's probably 4.5-5 times larger in physical size. And then to contrast it again, I have a bizarre Soviet core memory module out of a fighter jet of some kind - it's like a block construction, contains just insane numbers of hand-placed copper wires in a large 3-D structure, with massive amounts of square ferrite "cube" bits stacked in layers.. All developed within such a short timespan of each other. Fascinating stuff. Incidentally my Univac memory slice has the Pentagon marking, a marking only used on machines delivered to the actual Pentagon. *Edit:* Found the photo album I made of my collection: imgur.com/a/rvkUL

[@MichaelClark-uw7ex]

I had the honor of handling one of those old 3d core memory arrays, it was so cool, the ferrite cores were actually loose, they just floated/sat at the intersections of the wires, it rattled when you moved it.

[@MsMarciax]

A real work of Electronics art you can actually see and appreciate. Would be a splendid idea to frame, wall mount as the heart and soul that went into these boards can be really felt. This has really made my day seeing it, thanks.

[@Mythricia1988]

I have a few of these myself, and the plan is indeed to frame them. I have one from the US (Univac, with a Pentagon stamp, so we know where that came from!), one from the ol' USSR, and another one made in Ireland of all places. They are beautiful things. You can see some artsy-fartsy photos I took of my collection here: imgur.com/a/rvkUL

[@Streamtronics]

I have a small bare russian core memory array but didn't succeed in getting the writing/sensing right. I've always wanted to build a "core memory demonstrator" with just a micro, LCD and the memory, being able to put in some data like "hello world" and have that written into the cores. Wave a magnet across the core memory to see the data getting corrupted etc...

[@UpcycleElectronics]

Streamtronics I was thinking some kind of desk trinket along those lines. Something like a name plate or any little trinket with lighting and motion that corrupts the data would be awesome. How about a clock that uses a magnet to reset the time every minute, and a MCU to refresh it?

[@JussiKilpelainen]

Streamtronics You might find my kit interesting www.tindie.com/products/kilpelaj/core-memory-shield-for-arduino

[@richfiles]

Oh. Wow. That's a thing that exists? Huh... That's a thing that exists. Neat.

[@darjiaethera]

Jussi Kilpeläinen Oh. My. Goodness. That is freaking amazing. Thank you so much for designing such a thing!! I will be purchasing this in the future just to have something neat to play around with!

[@darthvader5300]

By studying the most basic ferrite memory core design, several dozens of computer nerds were able to figure out on how to create the 3 dimensional physical geometric pattern for the wires and for the ferrite cores using an over lapping technique using the basic CMOS principle for a proposed thin film transistor technology modified to act as a base for the original basic ferrite core memory which was started in 2011 and they were successful. Mass production on a micro scale was done successfully in the same year and they already have ironed out all of the kinks. What they have now in 2019 is a stockpile of thin film transistor based-ferrite core memory which is far easier to mass produce despite the fact it is more bulky, cumbersome, large size like a paper towel with a storage capacity of 2 terabytes. Being solid state, without having any moving parts to wear out, and it's flat ferrite core wafers can store memory data and information indefinitely without being affected by EMP, EMI, EMF, CME, Solar Flares, all data stored in them are safe from being damaged or destroyed.

[@ChrisPoach]

Please get some magnetic viewing film to show the magnetic fields.

[@EEVblog2]

I tried that, but either they aren't meganetised, or as a I suspect the field is just way too small to register.

[@ChrisPoach]

That's disappointing :/ Thanks for the reply!

[@Eletronicafg]

Wouldn't the magnetic flux be confined inside de toroid? It's a closed magnetic circuit, without any air gap.

[@stargazer7644]

They're toroids. The majority of the magnetic field stays in the ring.

[@eichwo]

If it is from an ancient Siemens Mainframe I can get you in touch with a retired Technician...Ping me if you want contact Details

[@WingoJuny]

Now, at 74, I remember my time maintaining an RCA110 computer that had a 4K ferrite memory. This was around 1968. When it failed, I disassembled it and managed to find a broken ferrite ring on a parity bit. This particular systems 'word' was 23bit w/parity. What a challenge, what a memory!.. You are correct in that the data was destroyed when 'read' but the machines of the time always did a 'read/right' cycle to retrieve data, thus maintaining the memory. Now all you need is a spool or ten of some paper-tape programming.

[@VauxhallViva1975]

Magnetic Core Memory is facinating stuff to look at. How far we have come.

[@charlesparmele]

In the US Navy, I worked on a computer for the Trident C-4 Navigation system that used core memory. The thing I hated most about core memory was making sure the currents were adjusted properly because if they weren't you would get spurious data. You would think it would be a simple procedure but it's not, you had to get everything perfect.

[@AZOffRoadster]

Back when I was in college (early '80s) I was given 2 old terminals. Sounded like a hurricane when you powered them up. When I tore them down, they had 2k of that ram in each. The keyboard was comprised of individual hall effect switches. If I hadn't parted them out, they could have flawlessly run continuously to date. Built like tanks to last forever.

[@BassGuitarGuy128]

A link to this video was included in the powerpoint from my computer organization class. Great video.

[@NathanChisholm041]

I Knew a old women who back in her day use to weave these cores with the wires! Sum job!

[@PCNeuverdrahter]

Soo, the german label says: Core memory calibration: 400 Words, Date 1970-08-30, Name C413 The columns are labeled "lower limit", "average", "upper limit" for the LEFT and RIGHT bits. no repairs made, only annotation a check mark ;)

[@gopro2027]

This kind of makes sense with george s's comment "I took Dave's bit bucket challenge and counted 'em. Each of the 4 sections has 100 by 64 wires, or 100 64-bit words, equivalent to 800 bytes per section, 3200 bytes total ." because since there are 4 of them, 4 * 800 byte (400 words) = 3200 bytes.

[@rnb250]

Wunderbar

[@cybermaus]

Please note: on modern DRAM, we *still* destroy the data when reading (the tiny cell capacitor gets discharged). We don't notice, because the chip has a line buffer and writes back the data before reading another line.

[@GregSr]

Back in 1980, I was a field service engineer. The computer running our machines was a DEC PDP 11-04. The thing I remember most about that computer was how hot it ran. The 8K core memory used a lot of current resulting in a lot of heat. I also remember that replacing a core memory board was VERY expensive. One of the common problems in those computers was the air filter behind the front panel. Most people didn't even know it was there. I'd pull the foam filter out and it was completely clogged with dirt.

[@QLTD]

it would be fun to try to use it with a modern electronic build

[@bobbymanganaro]

I know the very basics of how a computer actually works. Amazing Humans can come up with this stuff. Imagine what we do if we really worked together

[@alwilson6471]

I used to work on these machines back then - for Sperry Univac. Brings back memories working on systems so big you could have your tea break inside them! It's where I played my first computer game - Star Trek on a console printout.

[@LandNfan]

I started in computer programming in late 1974. The first system I worked on was an IBM System/3 Model 10 which had 24K of ferrite core memory. The OS took up 4K, leaving the other 20K for user programs. We were the odd birds since we programmed in COBOL rather than the native RPG II. You could write a pretty large COBOL program in that 20K as the linker broke it up into many overlays. Once I wrote a program so large it would not fit in memory. I finally hammered it in by replacing all of my carefully thought out, user friendly error message text with cryptic 3-digit codes.

[@ricknoah9184]

Ah yes, also remember having to make the actual programs rarely exceed 16k. You leaned to write very tight code. Not like today, by any means. Finally along came 640k RAM and it was like a whole new world. All that ROOM !! LOL.... but then, no one writes in machine language anymore, either. I could write in Z80 and 6809 and 6502 code all day. (We don't need no stinkin' compilers...LOL) Was nice when "C" was invented.. Yes, alas, I really am old... {sigh}

[@thomasgarrison7622]

When I first started working with computers it was on an IBM 360 mod 25, with one 16K module of core memory. The module was a 13 inch cube. This was in the late 1960s.

[@tronixfix]

Those 70's electronics are just pur art!

[@mikeselectricstuff]

That is not core rope. Core rope is a read-only system, famously used on Apollo, made by selectively passing wires though cores to select the data value, and the whole thing coiled up,

[@Sixta16]

Exactly!

[@sonofeloah]

My dad was working for uncle scam back in the 50s in the bell labs and was working with some others in a group that first developed these ferrite core memory "webs" (as they called them back then). Awesome to see this one intact. Would be much more awesome if you can get it up and running for us to witness. Read some good ideas on how to do that from other comments. Wish my dad was alive to see this.

[@darkstarnh]

In about 1980 I was working in the studios of HTV Television in Cardiff, Wales (part of the ITV network). The studio lighting system had core memory systems to remember lighting conditions for different shows. We thought that they were pretty antique even then!

[@RME76048]

3:03 an elaborate machine winding core memory = very skilled women sewing ferrite cores. I don't know that a machine was ever designed to stitch core. Please correct me if I am wrong on that, folks. 3:56 I believe what you get is what is stored in the ferrite torroid: the magnetic field is circulating either clockwise or counter-clockwise so you get either a positive-going pulse or a negative-going pulse (not a positive pulse or no pulse at all) which you could define as representing a one or a zero. Either way, the reading of the bit state was destructive and was followed with a rewrite of the bit.

[@andyj2106]

If you want to see some Core Store memory units, visit The National Museum of Computing in Bletchley, UK. You can also see machines running that use it and talk to experts about it!

[@mikus4242]

Interesting tidbit on magnetic core memory: Since it retained its values without power, if a program or the OS crashed, they would run a "core dump" routine to read out exactly what the computer was doing at the time of the crash. They could reconstruct what caused the crash. When I was studying computer science in the 80's, we would refer to a test where we had to write a lot of information down as a "core dump."

[@stargazer7644]

It is still called that, even long after computers no longer have core memory.

[@Edelweiss-uv5xi]

It wasn't elaborate machines that did it; it was all woven by hand.

[@TVPiles]

Siemens had a book out around that time with instructions on how to make those thingies by hand and how to troubleshoot the logic (Digitalspeicher mit Ferritkern, G.Keller, G. Pumpe & Rolf Kramer, 1971) . It proved invaluable to me when I build my first computer (1973).

[@user-yb5cn3np5q]

Finally someone properly reads the word "data".

[@chromabotia]

I remember ring core memory. I was told that the ring core I was working with was hand assembled by women at Ampex in Redwood City, Cali

[@napalm5]

Reading data using the sense line also erased it if it was in the opposite polarity. The data was written back to one by counter intuitively writing to all but then inhibiting all but one during write.

[@knusern666]

I got a look at a ferrite core board when i was in Norwegian army in 1995, it was used in the mobile artilery as a targeting computer. According to the officer that had it on his shelf ( it was no longer in use by that time) it was hand made.

[@laharl2k]

Dave, you could be the first person in the world to store porn on a magnetic core memory!!!!!! DONT MISS THE OPPORTUNITY!

[@EEVblog2]

Should a I store an ASCII dick'n'balls?

[@DIYTAO]

How about bitmaps.. each of those blocks seem to have 100*64 bits of data.. With 4 you could make animation :-)

[@LemoUtan]

Pretty sure he won't be the first. Pret-ty pret-ty sure.

[@laharl2k]

Some ASCII art or a 100x100x4bits .jpg (seems to compress better at super low settings than GIF) could probably fit, though ascii art would probably look better

[@MichaelClark-uw7ex]

Nah, those Seimens guys probably already did it in 1970.

[@mbaker335]

I could be mistaken but I thought the space shuttle used some of this memory as old technology is favoured for space flight. The z80 and 6809 were used in space probes until quite recently. All the bugs are known and the devices are more robust with larger structures.

[@Mythricia1988]

Another reason they kept using the core memory for the Shuttle is that core memory is almost completely impervious to particle strikes and high radiation environments, where as semiconductors have to be radiation-hardened, which is expensive, and can affect the performance characteristics of the devices.

[@shadowkat1010]

I believe that it was cheaper to have them manually wound by humans than create machines to make core memory, so chances are it was built by humans.

[@fredfarnackle5455]

I worked as a senior toolroom technician at IBM Havant, UK in 1970. Part of my job was to maintain (and fix) the relay driven machines that welded the X and Y wires to the memory backplane. The first step was the cores were loaded into a fixture in the correct orientation by a vibrating machine and held there by vacuum, then they were manually inspected by a bloke with a high powered microscope - any fault at all and the whole thing was junked and later re-loaded with fresh cores. After that, the loaded fixtures were carefully placed on the welding machine and vacuum applied again, then the operator fed all the X wires through by hand using a wide roller (not as easy as it sounds!), the correct tension was applied and they were welded and cut off all at once when he hit the weld button. The fixture was then rotated 90 degrees and the Y wires were done the same way. Then they went for more microscope inspection and if OK went to Portugal where women fed the sense wires through by hand. On return to UK they were all manually inspected again before electronic tests. Those relay driven machines were a bugger to fix, they were that old. I have an old core memory board from an unknown computer, it is 4.5k bytes and I think is older than the one in this video as it has no IC's at all, all signals were routed around by diodes: I believe a similar one would have cost around US$36,000 in the 50's, dropping to around US$360 in early 70's after methods of mechanical assembly were perfected.

[@annother3350]

Knitted by Grandmas!

[@zadrik1337]

I knew a researcher who had a Dec PDP 8 in the antarctic. He said that They would power it down when they left for the winter and power it back up when they got back and the core memory was still at the same values. They didn't have to boot the computer, just toggle the run/pause switch. Awesome stuff even if it is bulky as anything. There are some really good core memory training videos done by the army. Just search for "Electronics: Magnetic Cores I: Properties 1961 US Army Training Film"

[@willgeorge5644]

Just as well we did not have to re-load the memory, we had to load it from paper teletype machine tape. took hours to load, had to load the boot louder program first, manually, bit by bit! Set up the instruction on a bank of 32 switches, then toggle the "load" switch, program counter automatically goes to next location, repeat with next instruction. We learnt to program like that, understanding what every bit did!

[@moeshouse575]

many years ago i found a book about ALL of the Magnetic Core Memory and the diff vers and how they was made. the ONLY book that i read EVERY word of.

[@JohnAudioTech]

Interesting seeing integrated circuits on the same board. I don't know the state of memory on monolithic silicon back in 1970 but it had to have been far more compact.

[@Mythricia1988]

More compact maybe, but also much more delicate, much less proven, much more expensive, and for not much gain (it was quite slow compared to modern semiconductor memory, and also, core memory became faster and more compact as time went on). Core memory was produced well into the 80's, and reach pretty serious density levels.

[@graemedavidson499]

There's probably still data stored on there - would be interesting to read it out to look for any plain text!

[@johnnyprimavera2]

Might say that the golden DIPs at 4:15 are SGS-Thomson (ST) Octal Bus transceivers from the 74 series logic.

[@olafzijnbuis]

The movie from the development of the AGC (guidance computer for the Apollo project) shows so-called memory. Here many wires go through or outside a core. It was Read Only Memory. See: pixabay.com/en/computer-magnetic-core-rom-pcb-1346042/

[@MichaelLloyd]

Interesting. I think I was in grade school in 1970 and didn't touch my first PC until I got my hands on a beast of an IBM PC sometime in the mid 1980's. It had two 5-1/4" floppy drives and an 8086 processor WITH a math coprocessor. Soon after we updated it and replaced one of the floppy drives with a monstrous 10Mb hard drive. DOS 2.0 had just arrived on the scene.

[@MichaelLloyd]

Yeah... AND 256k of RAM...

[@trueriver1950]

I worked on a miniicomputer with 128k of 8 bits in ferrite rings. Reading out the memory destroyed it as you say. However equally bizarre by todays standards is the fact that powering down did not lose memory. We stopped our machine in Slough Uk, shipped it to Anarctica, turned it on after 6 months power down and all our programs were intact. Pressed run and it did. No need to hibernate - core is inherently non volatile on power down. The 128 kb took about a foot of a standard 19" rack. Those Perkin Elmer 8/16 E minis were last manufactured in 1980 and may have been the last gasp of core memory. Turning to your unit - not only might you be able to store data but it almost certainly still contains data from when it was last powered up. So please do a read before you write.

[@OrionFyre]

I remember in my computer science class in '96 building a small array of core memory like this. Nothing so huge, probably only 64 bits at the most. It was a good project to bring the abstract thought of bits and bytes into reality while doing some communication out of the computer.

[@intel386DX]

this modul is cut version , in the middle there is a space for more memory of core :D

[@PlasmaHH]

You could flip certain bits there and have a look with the thermal camera...

[@bryanlatimer-davies1222]

In the 70's I repaired NSM juke boxes that used a very similar system to store the selections made, so being destructive on read meant the record only played once ideal

[@DasIllu]

My school required me to do an internship in a company of my choice. A friend of our family had a computer and tv repair shop and that's where i went. The guy gave me one of these. I was totally fascinated by the thing. I could ofc never expect to get it to work (i was 12 or 13). But just looking at it through a magnifying glass was awe instilling. Geee... that's almost 30 years ago.

[@Land-of-reason]

When I first started research for the UK MoD 1975 our computer system had two HP2100 plus an FFT box. The HP2100's had 16K and 32K of 16bit Ferrite core memory with a further 16K in the FFT box. I remember that the memory was expensive but hadn't realised that it was that expensive. At that time I understood that it was about £1K per 1K. I understood that it was all hand made. In terms of today's money 2022 that is £634,505.24.

[@GH-oi2jf]

A matrix of cores is generally one bit of a word, for as many words as there are cores in the matrix. It looks like you have four matrices, so four bits. You need to have as many modules as you need to make up a word out of 4-bit slices.

[@olafzijnbuis]

Cores were also used in calculators to store a numbers. Look at this 16 x 18 memory plane from a 1968 calculator. pixabay.com/en/core-memory-unit-calculator-3406734/

[@jonahansen]

This is where the term "core dump" came from - still in use in UNIX/LINUX derived systems.

[@feynthefallen]

I once saw an even tighter weave of ferrite core memory, I think it was from 1976. It's in the collection of the Institute for mathematical machines and data processing (vulgo "Informatics department") of Erlangen University. Last I saw it, it was on display in the blue tower, so check it out if you ever get to Erlangen in Germany.

[@Anamnesia]

I'm going to have to stop playing the *Dave Jones drinking game* ... It's too easy to get hammered, taking a shot every time Dave says, *Basically* ! 😮 🍺 😎

[@Zodliness]

Can you imagine the size of a portable camera, if that thing was the only available plugin memory?

[@musikekul]

They had an old magnetic core memory board on display when I visited the CERN computer centre.

[@WacKEDmaN]

LOL @ Daves nerdgasm face when he realises what it is!

[@terawattyear]

As a kid in the 60's and 70's I remember looking at a picture of memory like this in the "World Book Encyclopedia". Funny... I had always had the conception that the ferrites were much bigger (due to the image in the encyclopedia). Very cool to see a real setup. Thanks Dave and thanks to the viewer who shared the memory with you. I wonder how long a set bit would persist in this memory. Maybe you could read what is in this piece.

[@Mythricia1988]

Depending on the specifications, the expected retention period for ferrite core memory was anywhere from 10 to 100 years. Although, I think the chance is pretty high that you just erase everything when trying to read it, unless you do it just right the first time :P

[@WinrichNaujoks]

I was 9 days old when this was made! Cool piece of tech.

[@craigrotay3732]

Old memory is amazing. This is twistor memory. Telcos used piggyback twistor memories. Very cool. Thanks Dave, thanks to you first time I've seen the actual thing. I'll try to see if I can find datasheets on those controllers. I work for ST and joined when it was SGS-Thomson/MOSTEK

[@victorjohnson7512]

I worked on old 1960's era Navy Jets that were still in use in the early 1980's. They has the same magnetic ferrite core memory in the Nav-computers (80 lbs worth) about 16K memory...

[@NeoMorphUK]

The first computer I ever operated was an Elliott 803. That ran on core memory too and the memory core was frickin HUGE.

[@KarlBunker]

There's something incredibly beautiful about that board; it's like looking at skillfully made neolithic stone axe.

[@rich1051414]

There was no automation. It was all the scientists mothers and grandmothers brought in to weave them by hand. They were quite proud to do it, I am sure. The programmers called this LOL memory, for Little old lady memory :)

[@joevignolor4u949]

Little old ladies enjoy knitting - same thing.

[@rich1051414]

@@joevignolor4u949 Well, it was a matter of hand size. You needed small hands to do it, and women generally have small hands, and grandmothers usually needed something to pass the time anyway.

[@LydellAaron]

This is still cool. A board like that with some ferrite cores that could light up or something, so you could see them activated or not, would be a very helpful learning tool. The destructive part of reading back of value...interesting...can we use that accidental somehow to model the action at distance phenomenon from quantum mechanics? This board size, would be the perfect board size for a version 1 home Quantum computer today (2022).

[@GeoffreyFeldmanMA]

The sense wires were not automated. They were sewn in by hand. While in college in 1971, I toured an IBM facility and saw core memories being made. The cores would be shaken out over a template board where they mostly would settle in where they should be. Then the few that hadn't settled would be hand placed. A loom would then push through the X and Y. Then the sense lines would be sewn in. I also saw testers with the assembled stack of core planes, a cube about 1'x1'. I remember seeing some sort of test signal going through while the tester would tap the cube with a rubber mallet and observe the stability of the sense signal on an oscilloscope. In terms of automation, by modern standards, not many of these were made. This was not even close to being a consumer product.

[@cadbury204]

Not to be confused with Core Rope Memory, which was used on Apollo. en.wikipedia.org/wiki/Core_rope_memory

[@Bengadeer]

May be a bit out of date(sorry about the pun), but fun to see. I date back to those days myself. In fact, the Tomahawk cruise missile had a core memory in its guidance set! Better yet are the drum drives that would explode occasionally when a bearing failed, and they would, because they spun as fast as mechanically possible. The drum case was 1/2 inch thick aluminum to hold the fragments after a failure!

[@typograf62]

I have another board of core memory, roughly the same size (a little larger). It is marked as from a DGC Nova from 1972 but may have been used in a Danish computer. It has 16 K words of 16 bit so the cores are really small.

[@SoundzRite]

Back in my 1974 Digital Logic Class where we were using a DEC PDP8 in lab, my prof was showing us the core from one and explained that it cost several thousand dollars because it had to be made by hand. As he was handing it to the first student to pass it around, he horrified us all by clumsily dropping it. After a chuckle, he explained that the PDP8 had gone to semiconductor memory and that core was pretty much worthless at this point.

[@fusseldieb]

I would love seeing you storing something on it. *Absolutely!*

[@ironman7261]

bought one about a quarter the size of one side from jameco for 6 dollars back in 1974 still have it

[@Big_Tex]

Awesome to see this for real, I'd only ever seen drawn diagrams of these things. Had no idea what size the cores really were.

[@electronash]

Modern DRAM uses "destructive" reads, too. (but obviously based on microscopic capacitors on silicon gates, rather than magnetics.) They normally have a Row Buffer, so the data being read gets written back on the next cycle. There's an interesting security hole due to that process, and you can still even get "crosstalk" between adjacent rows with DRAM as well... kzbin.info/www/bejne/ooS4Z4ywn658n7M

[@electronash]

And of course, Jeri did a vid about similar tech a long time ago. kzbin.info/www/bejne/pmi2nHhrpaqFqqM I would have linked to xjet's vid too, but he's already on here. lol

[@rich1051414]

I thought DRAM was destructive whether it is read or not, and the memory has to continually be written back.

[@electronash]

Yes, that too. ;) The charge leaks out of the capacitive cells over time as well, so they have to be refreshed at a regular interval. A common refresh rate is 64ms, but it's obviously impractical to do this on all cells / rows at once, so it just refreshes every row in sequence. ie. that whole 64ms is split up into how ever many rows there are, so if there are 1024 rows, then the timing between each row being updated would be 0.0625 ms (62.5 microseconds). So, eventually all the rows will still get refreshed, and an individual cell will wait no longer than 64 milliseconds before it's next refresh. ;) (It took me a while to realise how the "auto-refresh" in DRAM worked when I first started messing with computers / electronics. All it really has is an internal counter that increments, to refresh each row in turn. At least that's the common method.)

[@integraxnine745]

I was looking for such a video for a long time. I was interested since Jeri Elssworth mentioned them!.

[@brianr987]

Nonvolatile memory. No power still maintains state.

[@ghostrebel2336]

@Shufei Ypu forgot to turn off ECHO

[@willgeorge5644]

I worked on a postal automation system that used an ITT computer with 32K X 32bit core memory in 1976 - 8. That computer was new then. I still have some cores somewhere, not from that computer, they made wonderful drive transformers for switch mode PSU's due to the square magnetic loop.

[@Mr.Unacceptable]

You put Linux on what?

[@gaxnn]

I worked on old systems that had core memory in the early 80's and I also worked on systems with delay lines. Interesting fact about core memory, if you had an error in your program that caused a tight loop (like an instruction that branches to itself), you could actually burn out the core memory drivers. Also, with the delay lines, they came mounted as a long coil of wire in a case the size of a hardback book and if you hit the case it would corrupt the data.

[@edgeeffect]

ICs and core-memory on one board... that's an important historical artefact!

[@JustWasted3HoursHere]

For those interested, the story of the Apollo navigation computer, which used a similar "rope memory", can be watched here: kzbin.info/www/bejne/roKUgGOOh8SWd5I The entire "Moon Machines" series is HIGHLY recommended. They stored the program code with it, so in the case of rope memory, the value is not erased when read (unless it is re-written with the same value after it is read).