Motorola 68030 wire wrap prototype | Hacker News Reader

Motorola 68030 wire wrap prototype

retrocomputingforum.com

189 pointsbcrl70 comments

Ha, I (like the commenter on the post) thought this was going to be some insane modern hobby project until I clicked through. It's in fact a set of period photos of the original (industrial) prototype. Very cool either way.

Some context for the unfamiliar: The 68030 is a historical 32-bit CPU from the Motorola 68000 series, following the original 68000, the 68010, the 68020, and their various variants. Besides a bunch of 80s Unix workstations and pre-PPC Macs, it was used in several later/higher-end Commodore Amigas and in Atari's final computer, the Atari Falcon. Unlike the 1MHz prototype in the fine article, the production chips ultimately ran from ~16MHz to ~50MHz, as I recall. Comparable in processing power and historical era to Intel's i386, as another commenter said.

Actually the '030 was sort of a flop in the workstation market. NeXT used them, but all the big Unix vendors swung hard to RISC right as the '030 reached market. The iconic 68k workstation was and remains a Sun-3 68020 box with Sun's own MMU.

And while it's true that at release the 68030 was very competitive with Intel (but not SPARC or MIPS), by the release of the 486 two years later they'd lose that crown and never get it back. Their window turned out to be too small.

The G3 was competitive, wasnt it?

Benchmarks at the time claimed it was 2x faster than the PII.

https://www.baltimoresun.com/news/bs-xpm-1998-04-27-19981171...

So true or not, it must have at least been in the ballpark!

As I Mac user at the time, I remember the G3 providing a measurable performance leap!

G3 was PowerPC though, not 68k. ARM is actually the fourth Mac ISA, after 68k, PowerPC and x86!

You are correct of course!

I should remember really, I worked at Apple in the UK during the 90s! :-)

Guess I'm getting old LOL

I think my mind was getting it's memory mixed up with the PowerPC 601 and 603!

Too many product codes in my brain! :-)

Yeah, though Sun did do one 68030 workstation that I know of, the Sun 3/80, at the very end. It was packaged in the same distinctive chassis design as the SPARCstation 1 (Sun 4/60). I don't know the story, so I could only guess it might've been for hedging, or for customers still tied to m68k.

Yeah...it was pretty much an olive branch to the dead-enders who hadn't drunk the RISC coolaide. Most all of the shops I knew who bought them had some critical app that wasn't being ported to SPARC. Sun also had the 3/400 series servers that were 68030 based.

The 3/80 was also notable for using the Sparcstation expansion bus called 'SBus', which Sun used to claim SBus was an 'architecture neutral' expansion bus and tried to position it as a competitor to PCI (for a very short period of time). I went to a Sun SBus hardware design class at one point...it was a nifty design but clearly wasn't getting market traction outside Sun.

I didn't know about the 3/400 series.

One nice thing about the migration to SPARC is I was able to personally buy a retired 3/75 from the local Sun field office as a teen. Unix and megapixel display for a fraction of what assembling that with a PC would've cost. :)

HP was a big Unix vendor, but they were also vendors of other OS such as DOMAIN, and HP kept the m68k alive right through the '040s with Apollo-derived workstations that became part of the HP9000 line. DOMAIN/OS died out with PA-RISC and the 9000s segued into a new architecture, but not before cementing a legacy of a really well-done distributed network OS, a good way for college kids like me to learn assembly, and a cool OpenBSD platform for me in 1999.

I actually traveled the country replacing TI DNOS machines (running on the 9900 CPU).

The second phase of these replacements were with TI 1500 servers, running various 68k processors on nubus.

https://www.ricomputermuseum.org/collections-gallery/equipme...

We also should not forget the Sega Genesis.

> We also should not forget the Sega Genesis.

Thank you - I forgot about the 68000-based consoles :) Ironic, given how many great articles about console architecture[0] have done the rounds recently! The Atari Jaguar and Sega Saturn also had 68000s among the many processors in their complex architecture, albeit explicitly as a sound coprocessor in the Saturn.

Edit: I guess I didn't exactly forget - I was restricting my mental search to the 68030 specifically. But all the 680x0 systems were cool :)

[0] e.g. https://www.copetti.org/writings/consoles/mega-drive-genesis...

https://en.m.wikipedia.org/wiki/Motorola_68060

The 060 challenged in some ways, and was used on Amigas in the end days. Hybrid 060 + PPC boards existed too.

I don't think it was a flop for technical reasons so much as Motorola had spent the last few years basically telling people to stop using them and move to 88k -- -- and when that failed, to PowerPC.

Motorola could have done a Pentium competitor in the 68k architecture, but did not (well, ok, there was the '060 but it was half-hearted). They killed the line. And along with it anybody who had stayed on that ISA (Commodore, Atari). Though somewhat revived it later in the ColdFire, but only really for embedded.

The 68060 was anything but half hearted. It was one of the first commodity superscalar processors and in the end had a long life in the embedded space. But, yeah, the writing was on the wall that the 88k was the future, which to be fair is an excellent design, until it wasn't. Folks like Tektronix and DG who bet big on the 88k were not pleased.

The 060 was released[1] a full year after the Pentium and was significantly slower. It certainly wasn't a bad CPU, in an alternate universe where Motorola didn't fall as far behind on fab process as they did (or were able to find a second source to compete with Intel et. al.) you could absolutely see 68k continuing on as a dominant architecture. But in the real world of 1994, it was too little, too late, and is remembered now more as a bookend to an era than a processor design.

[1] On paper. In fact it never really "reached market" until significantly later when a few folks started shipping it in embedded contexts. The only desktop/server/bigsystem users ended up being upgrades to existing '040 designs.

This is comparable in transistor count to an 80386 (273k for the '030 vs 275k for the '386). Notable because a clickbaitingly-titled article yesterday caused people to ponder the size of an 80386 built with discrete components: https://news.ycombinator.com/item?id=33762543

Of course, this isn't discretes as discussed. I'm sure there's some discrete logic, but it's mostly PALs and SRAMs.

Did Intel make any wire wrap prototypes of the 80386? Or did they just iterate in silicon?

Reminds of the (older and much smaller) prototype of the Amiga chipset: https://cdn.arstechnica.net/wp-content/uploads/archive/journ...

Source (with even more interesting pictures): http://www.floodgap.com/retrobits/ckb/secret/lorraine.html

Very cool, but there is no way those boards are 24" x 24". As you can see in the photos, there are 24 ICs across the board (along the narrow width of the ICs). ICs are 0.3" DIPs, and worst case they are 0.5" center to center. Those boards are on the order of 12" x 12". Look at the folders on the shelf behind the cabinet -- those should give one the sense of scale.

I recall interviewing an engineer from Data General around 1990 and he described work he had recently done. Data General had designed a workstation around the 68040, but Motorola was way behind schedule. So DG designed a discrete board set that ran the 68040 instruction set and they shipped it. He said it was faster than the production 68040 as well (at that date, anyway).

Are you sure that was DG? I have a daughterboard from an Apollo workstation that has a 50mhz 68030, 68881 and a bunch of logic that mates to a 68040 socket.

I don’t think you could build a processor to run the 040 instruction set out of discrete logic without completely blowing your budget but you could build a 68040 bus adapter like Apollo did.

Now that you mention it, I'm not confident. It was a one hour interview 30+ years ago, but that salient points are it was an east coast minicomputer maker. Apollo would be more likely now that you mention it.

I’d like to see an M2 made out of wire wrap! How big would it be?! How fast could it be clocked? How much would it weigh? How much power would it draw?!

Inquiring (and lazy) minds want to know.

This wasn't entirely out of discrete components; they had PALs, presumably representing fairly large sections of the processor's internal subsections.

The modern equivalent is simulation on a series of big FPGA, probably this one https://www.xilinx.com/products/silicon-devices/fpga/virtex-...

Presumably they don’t create the hardware in FPGA and just simulate the thing in software?

Simulation in software is much slower than running the design in FPGAs.

M2 is 20 billion transistors.

You'll get different answers depending on whether you're doing this as transistors or 74 series logic or whatever, but if we take discrete transistors and jam them in at 4 per square centimeter, that works out as 5bn cm^2 or half a square kilometer.

Frequency will then be limited by propagation delay across a square of that size. Maybe a few hundred kilohertz? Power .. megawatts? There's going to be a visible heat haze shimmering over your half square kilometer of transistors.

That is a fantastic image. Thanks!

I tried to play with "manual" wire wrap tool. It's easy to buy thin wires for wrapping but it was extremely difficult to buy wire-wrap ready IC sockets. The difference is in the shape of the socket pins, the wire-wrap requires long rectangular pins but such sockets are extremely rare and expensive. So my adventure with wrapping ended rather quickly.

I had a box-full until recently that I shamefully threw out :) Yeah, the pins were also quite strong (harder to snip with side cutters iirc), and square (or hex) because the wire needs to bite into the edge of the pin. You could get away with as little as three turns. They were long because you might typically make 3 or 4 connections per pin, so the height would soon stack up. It is really hard, but slightly possible to wire-wrap directly to the pins of old through-hole TTL logic if you have only one connection per pin. The main problem is that unless you use very fine wire and go gently the torsion will twist the IC pins off.

It's also possible to use a mixed method of wrap and solder to make point-to-point simple digital circuits.

It's too bad you missed out on the experience. It was a lot of fun, a sort of mixture of knitting and following a detailed map.

I was doing it with circuit designs I had made myself, so I had the fun of that early stage of deciding what to hook up, where to place things on the board, and so forth. I laid it all out on paper, then made pin-connection tables (again, by hand) and checked everything over and over.

Then it was on to the exciting stage of placing components on the board and wiring. To avoid confusion, it helped to lay the wires out very precisely, so all steps had to be done with care and frequent checking. Precision work like that is very absorbing. Time just goes by as you live in the moment -- a bit like getting lost in music or art.

Beyond this relaxation-through-concentration element, the actual materials were beautiful. Gold posts, and silvery wires, each gleaming in the bright light I needed to see my work taking shape. The wrapping itself took a little bit of skill, and it was lovely to feel the coils going on, each precisely the same as another. Even stripping the insulation was fun, somehow.

I had an electric wrapper, but I really preferred doing it by hand. More zen.

All of this was in the 1970s, and was aimed at purpose-built scientific equipment in a university environment. All the materials are long-gone and I have no photographs of anything. Everything is lost to history, except for isolated memories of the great fun I had. And, bonus, it all worked in the end!

We used that technique in my university in the late 90s in the computer architecture class. I personally found it quite boring.

I had the same experience. The lack of socket availability made it mostly a non-go. Which is a shame because it would fill the gap between breadboarding and PCB for many DIY projects.

There's an alternative method using regular IC sockets and header pins - works well:

https://github.com/linker3000/Z80-Board

Get square male pin headers (cheap!) and solder bridge them to the pins of a regular socket in adjacent perfboard holes. I love wire wrapping for home projects but I've never in my life seen a proper wire-wrap IC socket. Square male pin headers are the right size for wire wrapping.

I remember when that came out. The Batwing was so greedy they priced themselves out of the race to give the parts distributors a load of easy $$. Sadly the price stopped all the builder/proto EE's from buying them and the corporate $$ was not enough to overcome Intel, and there was Gates's DOS and the CPM guy got into a bar fight(fatal) and the grand monopoly was born. The Apple guys did a good job bootstrapping Apple in a garage ( like HP) - even so IBM nearly crushed them except for the fact that Gates kept the right to sell DOS as well as IBM could. IBM tried to branch the ecosystem with their proprietary funny bus, Micro-Channel Bus. Zenith and a number of industry fools followed IBM down that rabbit hole - the earth fell in - they all died. Billions wasted buying selling stuff. Once MS Windows emerged, they all bit the dust. IBM even tried to fork Windows - badly - they all died too...

Is it correct to assume that "The Batwing" is a reference to the Motorola M-logo [1]? Might not be obvious to all so worth asking. :)

[1]: https://www.google.com/search?q=motorola+logo

Yes.

We called it 'Life under the Batwing'

Sounds like you're talking about earlier history; 68000, not 68030. This is 10 years later.

Off topic, but I’m always glad to see niche forums like this pop up. They’re a dying breed, but IMO a big part of what made the internet cool to begin with.

Wow, wire wrap. That could be a nightmare to debug, because the contact between the wrap and the post could be flaky. And if you had to change something, you often had to cut a wire, but you couldn't remove it, because there was too great a chance of disturbing something else in the rat's nest of wires. But that left obsolete wires with cut ends, and they could also on occasion make contact.

It's amazing that anything ever got working at all...

I remember around 1996 we made a prototype that was a single board about 14x12 inches IIRC and after a certain point of adding things, the digital stuff started behaving erratically. When we looked at the 5V DC line it had started approaching a sine wave shape from 2V-5V, so we switched out a lot of discrete chips for an FPGA and started over to get our prototype working again. I only did the software and wonder if we could have got it working with just wirewrap cause this 68030 seems like a lot more stuff.

Wirewrap is actually very reliable. From Wikipedia:

> Wire wrap construction can produce assemblies which are more reliable than printed circuits: connections are less prone to fail due to vibration or physical stresses on the base board, and the lack of solder precludes soldering faults such as corrosion, cold joints and dry joints. The connections themselves are firmer and have lower electrical resistance due to cold welding of the wire to the terminal post at the corners.

It was used on satellites until fairly recently.

Back in the 80s / 90s, I worked for a company that made custom compute boards for military applications. The boards had to withstand the vibration levels, temperature extremes and other stresses encountered in a main battle tank. They were all wire wrapped. Frequencies were low, though-about 25MHz max.

Yup, very reliable for low(ish) frequencies around 20MHz. I built 68010 and 68020 prototype systems using the DIL packages. Those were better for wire-wrap than the later pin grid versions of the 020 because you didn't get such a density of wires fanning out. A great text book we used circa 1990 was Alan Clements "Microprocessor Systems Design" (PWS Kent). It's an almost faultless book based on his practical systems lectures at Teeside Poly, full of schematics from which you can build working 68000 computers using pairs of 8 bit DRAM and EPROMS of the time.

This is interesting. I presume in the example shown they stuck to 1MHz as it would be more reliable than clocking to the kind of speed you mentioned, again presumably because the nature of a bunch of discrete components meant the switching of logic levels imposed a hard limit?

Another sibling mentioned the Amiga prototypes, I'm again presuming that the ~7MHz that that chipset ran at was well within the acceptable ranges for clock speed using discrete components?

<grabs book and dusts-off failing pages...> On page pp 142-3 is the minimal config schematic that shows "oscillator module" (IC4) but it isn't in the BOM. However pp 108 (fig 4.2) shows it as 8MHz. IIRC it was a MAX 6 pin 4MHz module I used, later I cranked it up to 10MHz on a 8MHz 68k uP and it still worked. Crazy overclocking!

Clements book is indeed excellent. I doubt I'd have been as successful homebrewing a 68k back in the day without it.

IBM's computers of the 1960s had wirewrapped backplanes and they were reliable. (The Xerox Alto and Apollo Guidance Computer are two other systems with wirewrapped backplanes.) I don't think we've ever had a backplane problem with the IBM 1401 at the Computer History Museum. IBM did a lot of field changes to the wirewrapped backplane, either to fix bugs or add new features. They had a specific process to go through. You couldn't just cut a wire; you had to unwrap the wire and remove it. But there could be up to 3 layers of wirewrap on one pin, so in worst case you had to remove 2 wires before you could get to the wire you actually wanted to remove. After you replaced the wire you wanted to replace, you'd need to replace the wires you removed to get to that wire.

And hopefully the other ends of those two were top-most, so it could end there. If not, there could be even further collateral rework.

Represent that network of wires as a directed graph.

Find the biggest tree spanned within the graph. This will be the wire that, should you ever have to replace it, will have you cancel all your other plans for some foreseeable time…

Yes, I think it was a requirement that both ends of a wire were at the same level on the pin. So you might have an additional top-level wire but you wouldn't end up with a long dependency chain.

I should also mention that the original (non-field) wire wrapping was done with an automated machine, the Gardner-Denver Automatic Wire-Wrap Machine. It had X-Y carriages with wire-wrap tools and it wire-wrapped the backplane according to a deck of punch cards. Details: http://www.bitsavers.org/pdf/gardnerDenver/SM506753_Paramete...

You had to follow a careful set of rules when wiring it up, but it worked-- for slow stuff. Anything faster than a few megahertz and the lack of proper transmission line traces will round off your edges, or give serious crosstalk if you drive them harder. We had to give up on wirewrap years before this.

We had a roundness/excentricity measuring machine at an old workplace (truck engine plant). I saw it get a full service+upgrade once. Electronics was all wire wrap. This was in 2005. Apparently they made so few of these machines, and each was made to order with custumizations so wire wrap made sense. And the service guy said that wire wrap was way more reliable than solder.

> the contact between the wrap and the post could be flaky

Not if done correctly. A wire wrap joint is more reliable than a solder joint, and much much more reliable than the socket connection sitting above it.

I don't recall seeing more than a handful of flaky WW connections ever from someone who knew what they were doing and using a decent wrapping tool. You saw them all the time from novices using cheap, crappy wrapping tools. That's undoubtedly where WW 'reputation' for unreliability comes from. I have stuff wrapped in the 80s that works just fine.

Anyone know more about the “Tiejen brothers”?

Don Tietjen and James Tietjen were brothers who worked at Motorola for a long time as I see their names in a 1979 document about the Design Philosphy of the 6809 and in Motorola patents as late as 1995.

> 18 24"x24" wirewrap breadboards

This is still in the realm of what someone can manage entirely in the head. I'd love to have a service which I could send an HDL of something around this complexity and get the chip on my doorstep. I'd like it also to have a reasonable cost.

If it's all digital logic, you can do this now with FPGAs. Modern ones would have no trouble implementing something like a 68030, and the tools can synthesize it from Verilog.

The great thing about FPGAs is that if you have a bug (which never happens to me, but I hear it happens to other people) you can fix it and recompile in minutes.

By hand? Wirewrap? That's going to be $1k per board!

For an actual chip, "shuttle service" does exist at even higher pricing.

The article describes the 68030 as a 68020 with an added MMU.

One of my Amigas a while back had a 68EC030, which is described as a 68030 without the MMU. So, I'm wondering what the difference really is between the 68EC030 and the 68020.

I too had an Amiga 4000/030 :) The 68030 added a minuscule cache (256+256) bytes, as well as more efficient bus.

slightly faster clock owing to the process shrink, and some changes to the bus. They doubled the cache by having a 256 byte cache for instruction and data separately. Instead of a unified 256 byte cache. Kind of a poor showing given they were released 3 years apart.

The 68030 was a 68020 leaving out a couple of rarely used instructions (I don't recall what) and adding a sort-of-but-not-exact subset of the 68851 MMU, 256 byte cache and some bus interface improvements (sync access). That and process improvements made it more efficient than the 68020. Motorola also upped the clock speeds on the 68030 (up to 50MHz).

The EC chips, at least at the beginning, had a defective MMU that was disabled and sold cheaper than a 'full' chip. Fine for the Amiga OS, or embedded stuff. I don't know if they ever made them without the MMU at all. The EC nomenclature extended to the 68040 and 68060, where it meant "no MMU or FPU", and added the LC nomenclature which means "MMU, but no FPU".

The software fad of the late 1970s was modular programming as seen in Ada or Modula-2. Intel supported that with its 286 and it was a key feature of the National 32016, so Motorola couldn't be left behind and added support in the 68020.

Except that by the time the 68020 came out Unix and C were king and C's dirty hack solution to modularity (#include .h files) was now considered good enough. Intel and National kept full backwards compatibility (to this day, in the case of Intel) but Motorola quietly dropped the stuff from the 68030.

This is interesting. It sounds more like a "simulator in hardware" than a strict circuit prototype. Was it used mostly to debug the microcode ahead of tape out?

Mind boggling.

Reply on news.ycombinator.com