back
29 comments
Followups since this was posted in November: http://www.bigmessowires.com/category/68katy/
God watches out for drunks, fools, and people who build multi-MHz logic circuits on solderless breadboards. :-P

From a signal-integrity perspective, he should not be able to get away with that.

He can't. "My 68008 runs at 2 MHz (it was unstable when tested at 4 MHz), providing similar performance to a 1 MHz 68000." He had to de-rate the clock speed of an 8MHz part to get it to run on that breadboard.

The 68000 line has a sad history. It came out in 1979, and for a while, it looked like it was going to be the winning microprocessor. Intel didn't have a 32-bit entry for years later. But early attempts to build workstation machine ran into problems. The 68000 had no matching MMU, and wouldn't work properly with an MMU anyway - page fault state saving wasn't quite right. Machines with external MMUs were built anyway, but great kludging was needed to get around the design flaw. The Apple Lisa avoided using increment instructions, and Apollo had two CPUs; one for the user and one for the kernel, with only one running at a time.

That was fixed in the 68010, which came out in 1982. But by then, IBM had picked the Intel 8088 for the IBM PC. Motorola didn't come out with the MC68451 MMU until a few years later, and it was a slow, segment-oriented MMU. So the Macintosh didn't have an MMU, and had a much weaker OS than the Lisa - no process isolation, no paging, not even a real CPU dispatcher. The UNIX workstation crowd (Apollo, Sun, HP, etc.) all developed their own MMUs for the 680x0 line.

If Motorola hadn't had those errors in the 68000 and had shipped an MMU earlier, the history of computing could have been quite different.

> That was fixed in the 68010, which came out in 1982. But by then, IBM had picked the Intel 8088 for the IBM PC.

Wait, that doesn't make sense - the 8088 didn't have an MMU either. Why would that have affected IBM's decision?

Looking back it does appear that Intel did nearly everything right, and their competitors didn't. It did help a lot though getting the IBM PC business, and due to that, aligning with Microsoft. If you look at the 68000 machines at the time they all had flaws that prevented them from really taking off, often not things that couldn't be corrected, but things that held them back. The Mac didn't have a hard disk, or multitasking, or color. The Amiga would always be constrained by the custom chips and that would be a problem long term, plus Commodore wasn't investing enough in software. Atari had a good simple product, and GEM wasn't too bad, but they didn't develop it further either. And the workstation market was constrained, as said by the parent, by the lack of a decent MMU solution. And then RISC came along too. In an alternative universe maybe the 68000 machines could have joined forces around a common unix platform and built a major market, but that didn't happen here.
What held back the 3b1? That's rarely mentioned in these parts, but it was a fantastic machine from a company that had experience building real computers (if even for its own use.)
There were a lot of early UNIX workstation companies in the early 1980s. Three Rivers, Apollo, HP, IBM, Apple, Sun, Sony, and AT&T all had 680x0-based machines. All totally incompatible.

At the time, it looked like lower-cost versions of those were the future of computing. But, as mentioned previously, Motorola was too slow getting the MMU situation fixed, which meant that all those workstations had some homebrew MMU that ran up the cost. Not until the 60030 in 1987 did Motorola offer an on-chip MMU. Workstations in that era cost $10K - $20K in 1980s dollars. Even the Apple Lisa was a $10K machine. Workstation prices didn't come down fast enough. Meanwhile, people were learning how to get things done on the original DOS PC, clones, and PC/AT, which had inferior technology but volume was driving down the price.

We had UNIX on the desktop in the 1980s, but few could afford it.

Seriously it's completely fine in the digital domain if the boards are relatively new, aren't crap (i.e. are Wisher/3M boards without those awful phosphor bronze contacts) and you're under about 10MHz and have enough decoupling capacitors and keep your signal paths relatively short. I've built pretty much this machine and it was frequency stable to about 8MHz and it only died after that because the clock was divided down from a 32MHz XO.

Stray capacitance isn't a mega problem unless you're doing analogue and require frequency stability or going above ~10MHz. The killer for digital is usually ringing on the high/low state transitions but if you look at the timing charts for most things this isn't that bad. Most digital stuff works pretty well if the signals are pretty fugly.

However, wave a scope or logic probe near it and all sorts of crazy shit can happen.

Was thinking the same thing. Especially with his "Schematics... forget it; organic creation" blurb. There's as much luck as wire in this build.

File it under "bet you can't do that twice in a row".

I've built commercial designs using that approach no problems at all. Most of my crap is built on PCB blanks and I have to spend some time reverse engineering it after I'm done.

Picture straight off my bench: http://i.imgur.com/AyXNHfX.jpg

I don't follow you. Dead-bug construction over bare copper lies at the very opposite end of the spectrum from solderless breadboarding. The two techniques have nothing in common.

You can use the dead-bug technique at frequencies well beyond 1 GHz with a bit of practice. The solderless breadboard, not so much.

I'm not talking about construction here; design approach. I don't have any schematics for what I'm working on; they get drawn up later when it's all working.

I have to reverse engineer deadbug and Manhattan stuff which is a downside which is my point.

Ah, gotcha. Same here.
Good lord
Google "jim Williams bench". I'm nothing :)
> fools, and people who build multi-MHz logic circuits on solderless breadboards

I think you may have just repeated yourself there.

I guess it's not really any scarier than 80s-era wire wrap prototyping, at the end of the day. Still, my hat's off to him just because he had the guts to try it!

He should send it to an FCC/CE test lab as an April Fool's joke.

In the 80s, we used Veroboard for that sort of prototyping. They would still have high-frequency problems given the huge amounts of additional wiring required, but I managed to get a Z80 board running at 3.5MHz. (Diagnostic slave board to a Sinclair Spectrum, so a bit simpler than the (very nice) board in the article).
Wire-wrap is actually quite good for durability, although it lacks an inherent ground plane. On the one occasion I had to send 20MHZ signals across wire-wrap, I had to make little twisted-pair lines by wrapping a ground wire around every signal wire.
Well, as I like to say: "because we can" is a awesome reason to do things! I still wouldn't want to try to debug such a system, though.
I think he is running it at 2MHz which should be doable.
I 'helped' a labmate troubleshoot a simple oscillator circuit in the 2 MHz range on a breadboard by turning my circuit on and off. Despite not designing any antenna-like features it was broadcasting enough to get his marginal circuit into oscillation.
Previous discussion: https://news.ycombinator.com/item?id=8622720
This is amazing, as a newbie in electrical engineering and just recently playing with my Arduino this is simply awesome!!
Reminds me of 18-545 at CMU. We did it with wire wrap though. Good times.

Congrats on being able to fit the kernel in 512K

Reminds me of the old PT-68K board; loved that thing, and the 68K.
Would mac os 9 apps compile and run on this system?
when ls -l /bin is being typed, all the last modified dates are "Jan 01 1970" why ?
Because they are read as '0' and that's the start of the unix epoch.