• Please review our updated Terms and Rules here

Why was there never a more "IBM-compatible" 80186?

Hak Foo

Experienced Member
Joined
Jul 18, 2021
Messages
213
It's well known that the 80186/80188/V40/V50 tended to be unpopular choices for PC clones because they offered "wrong" peripherals. Much of the value of integration is lost if you have to disable half the peripherals and wire in more compatible parts.

Of course, when the parts were designed, it's doubtful that anyone realized that a slapdash project out of Boca Raton was going to conquer most of desktop computing. Sure! Put in a GOOD DMA controller, it's not like everyone's going to use an 8237 and tape on the other four bits of the address!

But there was a good 5-year window between when the IBM clone market was legitimized (Compaq and Phoenix BIOS) and the death of XT-class machines as a mainstream commercial product. It seems like this would have been enough time to shuffle around some existing design components and bring out "186, but more IBM-compatible." It seems like an obvious product when your target market is either bargain-price clones, or early laptops where a "system-on-chip" design would save space and design complexity.

Did it exist and I missed it? Was the second-source ecosystem at the time not really capable of design rework, or contractually limited, so they could really only do naive-copy 186 parts?
 
It doesn't seem like "an obvious product" to me. What went into your calculus here?
 
The Jan 30, 1984 issue of Infoworld has a list of 80186 based computers, some of which were fairly IBM PC compatible. Lotus was of course one of the programs even the best 186 machines didn't get working. A little bit later and manufacturers could follow IBM and charge big bucks for the 286 or use the simple cheap yet fast NEC v-chips without working around all the oddities of the 186. The HP Hornet as seen in the HP100LX provided a tweaked 186 that could run most PC software so it might meet your requirements.
 
It doesn't seem like "an obvious product" to me. What went into your calculus here?

For the motherboard manufacturer, a design with only a few large ICs is likely to be easier to assemble and have fewer assembly-related faults (bad solder joints, unlucky pairings for corner-case parts). It might also be simpler to design for, if you don't have to place and route so many items.

For the chip designer, it locks more of the product's budget with them. The customer can no longer say "I'll buy CPUs from you, but (other brand) DMA controllers are a nickel less" if it's all in one package. The combined product might also be cheaper than the individual components.

I guess it might also make sense if you look at it from the other direction. There was an obvious momentum towards higher-integration chipsets. We got to things like the FE2010 which is basically everything but the CPU in a single package, so why not cross the last bridge? That basically gets you to the same place I was describing-- a 186-style "with integrated peripherals" chip where the choices are more IBM-clone friendly-- but from the opposite direction. I'm considering the integrated peripherals as the defining feature of the 186, more so than the new instructions.
 
The V40/V50 is basically what you’re asking for here, I’m not sure why you’re lumping it in with the original 80186. Its internal suite of peripherals has a few differences from the standard IBM PC setup, but it’s mostly only stuff a BIOS integrator needs to worry about, little of it is of consequence for most IBM compatible software. The 80186‘s internal differences were far more important. (The big deal breaker being the interrupt handling; The V40/50 can be set up to be a lot more compatible.)

Very late in the viable market window for ”Super XT”-class computers there were chips like the C&T F8680, which were basically an entire XT chipset, CPU, and SuperCGA card all on one chip, so what you’re describing in the end there did indeed exist.
 
The V40/50 might be closer, but they're still not drop-in replacement; you had to trap the attempts to manipulate the DMA controller that assumed 8237 behaviour, and likely needed a little custom bootstrapping on the BIOS to move the onboard peripherals to compatible places.

That F8680 is downright fascinating, C&T seemed to be a remarkably ambitious company.
 
The V40/50 might be closer, but they're still not drop-in replacement; you had to trap the attempts to manipulate the DMA controller that assumed 8237 behaviour, and likely needed a little custom bootstrapping on the BIOS to move the onboard peripherals to compatible places.

But in the grand scheme of things, not that big of a deal. Almost nothing that runs on XTs diddles directly with programming the DMA controller (I can speak from experience with my Tandy 1000 EX/HXes that you can have a highly compatible XT without having a DMA controller at all) and the end user likewise doesn’t care about BIOS initialization code. Obviously if you’re the one building the motherboard anything that replaces a handful of separate components with one isn’t a “drop-in replacement“, paying for a couple hours of programmer time is part of the job.
 
Also, bad timing. The 186 and 286 were developed in parallel, so the period where the 186 was the speed champion was pretty small. 186 and 188 chips were mostly intended for embedded applications, where they were very successful.
 
Another comment about the NEC V-chips, it may also be of some note that their entry to the US market came in under the cloud of a lawsuit which really didn't completely shake out until late 1986 or so, it's certainly *possible* it might have somewhat dampened the enthusiasm for the V40 (which didn't come out until about that time) among manufacturers that might be worried about offending Intel.

(Granted, "offending Intel" wasn't really something people worried about *too* much until the 386 came out and Intel tried to cut out second-sourcing.)

Honestly I suspect the bulk of the issue re: adoption, in desktop machines anyway, really mostly boiled down to the fact that by the time the V40 came along manufacturers had already started making their own integrated PC chipsets (The idea of using "megacells" of common support chips for cooking your own ASICs was really taking off around 1985 or so) and if you've already managed to skoosh your support chip count down to only one or two chips it really didn't make that much difference if the CPU still went in a separate socket. The separate CPUs were by this point dirt chip and available from multiple sources, so I doubt there was much pressure from anyone that was still churning out commodity desktop machines. At the very end there certainly was a niche for things like that C&T chipset or the Vadam VG230/330, which also included display controllers, for use in low power fixed-configuration PCs, but developing these chips wasn't really worth a mainstream CPU manufacturer's time, frankly.

Makes wonder why the Tandy 2000 went with 80186. Possibly supply issues?

In a world where strict PC hardware compatibility wasn't yet thought to be a make-or-break issue the 186 made perfect sense. The Tandy 2000 wasn't the only machine that made this mistake. (The Mindset is another famous/infamous example.) The IBM PC and the Tandy 2000 are at *least* as compatible with each other as your average two random CP/M machines are, that's how people still thought the MS-DOS market might work in 1983. They were wrong.
 
Makes wonder why the Tandy 2000 went with 80186. Possibly supply issues?
There were others, just not very common. This was in the days of "runs MSDOS" which was good enough. I was part of the Durango Poppy team--we used both the 186 and 286. The 186 was very attractive for a pure MSDOS system--integrated chip selects, expanded instruction set, interrupt, timer and DMA. You could design a workable PC with many fewer chips than an 8086 system.
Later variants of the 80186, namely that 80C186EC were much closer to the PC configuration datasheet, but by then it was too late.
 
The first 2000 that I ever saw in the wild was at the then Tandy Service Center in Traverse City, MI, maybe 1983 or so. I was there to get some warranty service on our office Model II and they had this 2000 setup for display. They were using a red plasma type display (not a fan of), and they had a spreadsheet up and running for demo purposes. Later on, when I was back with the feds, never saw another one or Tandy anything. Somehow I became impressed with the SX and decided that would be the one, but that's another story.
 
It doesn't seem like "an obvious product" to me. What went into your calculus here?
Vadem vg230 SOC is about as close as your gonna get.

Yes it is an obvious product a PC Compatible 186 or similar would cut the system cost in half due to the immense reduction in chip count.

The extreme popularity of the not IBM PC compatible 8018X shows it was obvious

Sadly just because an SOC is obvious doesn’t mean it would have been legal.

At the time only IBM would have been legally able to play and customize an Intel x86 compatible into an SOC

And it wouldn’t have to be an 80186, could be a 286 SOC or 8088soc , IBM had the tech at the time but zero willpower, they were too overburdened / distracted with mainframe design personnel in a collapsing market.

Intel at the time of the 80186 didn’t care about absolute IBM PC compatibility as they had their own “elegant and incompatible” solutions like the 8089, expecting them to do so would never happen, getting us to the F40 that released long after that obsolete x86 SOC was relevant released in the 486 era.
 
I recall that when we were looking for a 16-bit CPU (circa 1979), "Fast Eddie" our Intel sales guy mentioned the iAPX432 as "just what we wanted". Well, it got later and later and we'd already looked at the 8086 and wanted something better. Eventually, Eddie stopped talking about it and we started to express interest in the Moto 68K.
That didn't sit well with one BOD member, who was an Intel VP. So we were put on the alpha track of the 80186 and the 80286--multiple steppings of both. I recall one horrible two weeks of pounding my head on the floor when our test code kept crashing randomly. Since we were interrupt-driven with DMA for several devices, we considered the possibility that we messed up our driver logic. But hand-checking each line of code turned up nothing. If the system was sitting idle, it would run for days. Put a workload on it, and you were guaranteed a random crash.

Turned out to be a bug in the 186 microcode that apparently used the SI and DI registers for DMA--and forgot to restore them after a transfer--sometimes.

The misery with the early steppings of the 80286 made me wonder if protected memory would ever be made to work.

I still liked the 68K better, but it was going to be a cold day in Hades before we'd ever get past putting together a prototype. Top management wouldn't stand for it. NSC sent around a couple of guys periodically to talk about their 16032 CPU, but it was like asking about progress on commercial nuclear fusion. "soon" forever.
 
"Where did Motorola sell all those MC68K chips?" you may ask. Visit any "phone company" #5ESS installation and pull a random board off of their spares shelf. Unless you just happen to pull a "half grid board" or a "channel board", you have about an 80% chance of looking at one of the MC68K family running an SBC pretending to be a circuit board. Western Electric/Lucent/Alcatel used a whole bunch of them.
 
Intel at the time of the 80186 didn’t care about absolute IBM PC compatibility as they had their own “elegant and incompatible” solutions like the 8089, expecting them to do so would never happen, getting us to the F40 that released long after that obsolete x86 SOC was relevant released in the 486 era.
"At the time of the 80186" would be 1982 or before, so that's not an unsurprising argument. The PC ecosystem wasn't necessarily the centre of the x86 universe at that point, so it didn't necessarily attract special attention.

I sort of wonder when it did become the centre of the universe for Intel. The 186 and 286 were too early, but how much of the 386(DX) design was steered by the fact a lot of people just wanted a really, really fast DOS machine? If they figured they were selling to blank-slate designs with full 32-bit readiness, I'd expect it to look a bit more like the 80376, with some of the real-mode legacy cut off.

It's clear they were thinking PC-first by the time of the 386SX, because it's clearly product targeted for a specific niche in the PC hardware market-- teasing "you can have 386-enhanced Windows or OS/2" to people who would otherwise be served well by an equally clocked 286.
 
how much of the 386(DX) design was steered by the fact a lot of people just wanted a really, really fast DOS machine?
Well the virtual 86 mode as far as I know would fulfill this criteria.

As i understand it the 186/88 required you to re-implement most of the chipset to get PC compatibility. It may be that by the time this was done, it cost more than existing XT designs. Maybe the 186/88 aweful incompatibility in the tandy 2000 and others made further development a flop in terms of marketing.

I'm sure it would have been possible to build a PC/XT SoC. But early days chipsets were often multiple parts due to manufacturing defects.
 
It probably hit Intel the moment that the PC and MSDOS became the corporate choice for small computers.

I recall talking to an Intel application engineer about the 80386. His reaction was "We worked very hard to give you true 32-bit compatibility and what do you do? You run MSDOS on it." I suspect that Intel never expected the 80286 to spend most of its life in real mode either, given the one-way-ness (documented in any case) that switching to PM from real mode was simple, but not the reverse.
 
Back
Top