• Please review our updated Terms and Rules here

286 really that rare?

dfnr2

Experienced Member
Joined
Dec 6, 2011
Messages
466
Location
Dallas, TX
For an upcoming retrocomputing meetup, I thought I might install 16-bit Coherent on a 286 for show-and-tell. Leaving aside for the moment the logistics of obtaining a copy of 16-bit Coherent, I figured I'd snag a cheap 286 on Ebay. I figured there should be some kind of working machine for under $100. I was pretty surprised to find very few 286 machines at all, let alone under $100. Are these machines really so rare? I would not have expected a 286 to be harder to come by than an Apple 2, or more expensive than a working Apple II system. Am I missing something?
 
At the time, they were pretty common--until the 80386 came along. A fair number of 286 motherboards were replaced with 386 ones, as it was a simple upgrade that gave better performance and new features.
 
Supply and demand. For the most part any 286 around today is in someones collection. People are also looking for desktop AT cases to put their 486 systems into.

I would expect an IBM 5170 to command more then a cheap clone, but $100 is probably the low end for an untested 286 on Ebay these days, and then shipping can be expensive since most are boat anchors.
 
Remember that the number of computers sold more than double every year in 85-95 and when the 286 finally became the budget solution, cheap 386SX variants were available instead.

Given Coherent's preference for extra memory and the near impossibility to find 16-bit memory expansion cards, it would probably be easier to use an early 90s budget 386.
 
I figure that ebay itself is the biggest issue here. However, the 286 had a unfortunately life. They were very common long long time ago, and I was knee deep in them back in the day, but had very little use when obsolete. At best a fast early office machine, but not fast enough for Windows 3, too fast for the first PC games, too slow for most that came after. I had no use for them and a bunch stacked up. I got rid of them and none are left here. I wish I kept one of the better systems, or even still had my turbo XT, but storage space can only go so far, and moving usually is a deciding factor. I would think Apple II would have been as common in numbers, but not in households as late, and probably a bit more sought after. So between the market for these machines and obsolescence, time made these disappear largely before people got around to start respecting them again. I think if looking elsewhere than ebay you might find a decent box.
 
I think they just don't sell that well. For some reason the 286 has a reputation for being too fast for original PC and XT games, and too slow for 386/486 era stuff with good graphics. Most just think of them as good business computers. So perhaps the ebay sellers have given up. I've got a couple 286 systems and they are some of my favorite computers. New enough that you can find lots of 16bit ISA cards that work well in them, so network cards, IDE hard drive controllers, video cards, sound cards, etc are easy to come by, but they are old enough that you don't want to just load up windows 3.1 (but you can, it just is slow), and instead you can stick to older DOS programs.

Sometimes looking for specific models instead of 286 can help. Look for Compaq Deskpro 286e, IBM PS/2 Moodel 30 (most are only 8086, so have to check carefully), NEC Powermate, IBM 5170, etc. Some seller know nothing about computers and are too busy or too lazy to google what the specs are.

Being that you're in Dallas, you might try joining the Computer Reset Warehouse Liquidation group on Facebook and signing up to go to one of the events. (make sure to read the rules, they don't want 1,000 people signing up and asking what's there and if someone will ship them something, because no one really knows, it's not a functioning store, it's just a bunch of cool old stuff piled up to the roof. The group seems to be mainly to plan events and to show off what you picked up)

I'm not sure how much stuff is left that is as old as a 286, but it would be worth a shot.

There is also a computer Reset Buy sell group where you can probably find a local that picked up too many 286 and would be willing to sell one cheap to a local. In this group it's OK to ask if anyone has what you're looking for.
 
I worked in that era at a company where fairly speedy machines with lots of RAM were needed; the 286's weren't really that much faster than turbo XT's on common business software, and RAM was more important than raw speed (big Lotus 1-2-3 spreadsheets, for example). A good turbo XT with a good, fully stuffed EMS board could keep up and the considerable price differential between a good Turbo XT and a 286 wasn't worth it in many cases. Keep in mind that all of these machines were quite expensive and the idea of buying one as a home computer to play games with exclusively was very alien to all but very well-off people. Remember, no Internet, of course, and all you had were 9600 baud text BBS's or Compuserve (which was VERY expensive). An XT did just fine on a BBS or Compuserve. Business apps, Supercalc, Wordstar, Word Perfect, Lotus 1-2-3 ruled.

When I saw a Microsoft demo of "Windows 386" or some code name for it, I knew immediately that anything less than a 386 would be sent packing home as a "kid's computer" in a few years. MS made it very clear that "the cool stuff" that it could do would always require a 386.

I had the company's first 386, a Gateway. After my boss saw it and what it could do, within 3 months he and all of the engineers had 386's.

The "business machine/power user viability window" of the 286 was pretty short lived.
 
By the time 286 systems started becoming popular, 386es were already on the market. So people already knew there was something better; they were just waiting for it to come down in price. Plus most people were just using a 286 for real-mode DOS applications, so if you weren't in a hurry, then an 8088 would still suffice just fine. Finally, the release of Windows 3.1 in 1992 caused demand for anything less than a 386 to dry up very quickly.

And from a retro gaming perspective, if you switch a 386 or 486 into non-turbo mode, you usually get CPU speed similar to a 286, so there is no gap in the DOS gaming era that a real 286 is needed to fill.
 
It might be interesting to think in terms of "what would cause the machine to be saved?"

With something like an original XT-class machine, there were a lot of timing-sensitive programs that, frankly, the turbo switch wouldn't quite fix. You also had the fairly significant differences between XT and AT hardware on things like hard drive controllers, so you might be more prone to keep the old machine around in case you needed some data on it.

Conversely, there was relatively little a 286 could do that a 386 or 486 couldn't do just as well. There were a few pieces of software (some early Xenix and OS/2 releases, and a few utilities, as I gather) that touched undocumented or misdocumented stuff that ended up breaking on 386s, but most people never encountered them. So you probably didn't need to hold onto the 286 just in case.

I'd expect most 286 machines were being retired for upgrades in the 1992-1995 or so range. At that point, they held enough resale value that you're going to see people trying to unload them, rather than just sliding them in the attic and waiting for them to become vintage. I wonder if that led to lower survival rates: instead of first-owners saying "I spent $5000 on it, I can't come to terms with throwing it away", you had the people buying them second hand for $300 with far less attachment to it once they got a shiny new Packard Bell 486DX2/66.

Indeed, I could see price attachment being a big part of survival. I'd be unsurprised if the survival rate on "luxury" machines-- the AT, the PS/2 50 and 60, the Deskpro 286-- was a bit higher than on no-name clone 286s, particularly late-era ones where they were clearly budget offerings to fill the bottom of a catalog promoting 386 and 486 boxes.

I can recall my family selling their first PC (a Magnavox-branded 386SX/16) in 1995. We had at least one guy say "I refurbish the machines and sell them overseas", and he said for his purposes, he preferred a 286/20 or 25 because the 386SX wasn't meaningfully faster. I suspect if you actually wanted 386-native software, it would suck on a SX/16.
 
eBay prices have gotten stupidly insane in general the last few years.

That said, 286s in generic AT cases did tend to get upgraded to 386/486/Pentium 1 motherboards. You might need to buy a desired case and then hunt down a motherboard separately.

286s were awkward in other ways too. Early/cheap boards limited hard drives to a small list of supported drives in a ROM lookup table.

286s use flat XMS style memory - any support for using main memory as EMS had to be implemented in the motherboards chipset. Often they did not support EMS emulation or would require all memory over 640k be used as EMS in that mode. 386s could flexibly emulate that in software using the VDM support.
 
It might be interesting to think in terms of "what would cause the machine to be saved?"

With something like an original XT-class machine, there were a lot of timing-sensitive programs that, frankly, the turbo switch wouldn't quite fix. You also had the fairly significant differences between XT and AT hardware on things like hard drive controllers, so you might be more prone to keep the old machine around in case you needed some data on it.

Conversely, there was relatively little a 286 could do that a 386 or 486 couldn't do just as well. There were a few pieces of software (some early Xenix and OS/2 releases, and a few utilities, as I gather) that touched undocumented or misdocumented stuff that ended up breaking on 386s, but most people never encountered them. So you probably didn't need to hold onto the 286 just in case.

I'd expect most 286 machines were being retired for upgrades in the 1992-1995 or so range. At that point, they held enough resale value that you're going to see people trying to unload them, rather than just sliding them in the attic and waiting for them to become vintage. I wonder if that led to lower survival rates: instead of first-owners saying "I spent $5000 on it, I can't come to terms with throwing it away", you had the people buying them second hand for $300 with far less attachment to it once they got a shiny new Packard Bell 486DX2/66.

Indeed, I could see price attachment being a big part of survival. I'd be unsurprised if the survival rate on "luxury" machines-- the AT, the PS/2 50 and 60, the Deskpro 286-- was a bit higher than on no-name clone 286s, particularly late-era ones where they were clearly budget offerings to fill the bottom of a catalog promoting 386 and 486 boxes.

I can recall my family selling their first PC (a Magnavox-branded 386SX/16) in 1995. We had at least one guy say "I refurbish the machines and sell them overseas", and he said for his purposes, he preferred a 286/20 or 25 because the 386SX wasn't meaningfully faster. I suspect if you actually wanted 386-native software, it would suck on a SX/16.

Nice narrative and and puts a light on the times. My first "super" computer was a 386SX-40 in the early 90's. I paid $ 700 to the door for a mini-tower, motherboard with the 386. and the power supply - all known as a 'bare bones' setup back then. Then you had to pour over Computer Shopper to find some deal on the peripherals.
 
Arguably the '286 is more obsolete than even the lowly 8088. When it came out, it was an incremental upgrade. Faster, more memory. But the "286"ish parts weren't as amazing as they perhaps should have been, and the 386 really crushed it when it arrived. The 286 had really no value over the 386, even for the dollar, the 386 had much more potential.

In the end, the '286 lived a short life and everything since then has been 386+. At the same time, the '286 doesn't have quite the nostalgia value of the original PC, so it's sits in the middle, unloved and tossed aside.
 
The 286-20/25 would have been faster clock per clock then the 386SX and motherboards that supported those speeds probably used normal SIMM sockets so memory expansion was cheap and easy.

I have a few 286 machines in my collection but only one has SIMM sockets and a 286-16 or 20 in it. Most are 286-12 which was probably the most common one made. I even have a IBM PS/2 model 60 tower with a 286 in it.

Not sure why people are shitting all over the 286 chip. The 286 was around for quite a long time from 5 mhz to 25 mhz which overlapped the 386 era (just like people were buying high clocked 386 CPUs well into the 486 and maybe Pentium era). They spanned the MDA/CGA to VGA era .They were fine with DOS era programs, allowed the use of a FPU which you could run at a different speed then the CPU, and there were 16 bit memory upgrade boards for people doing serious work.

Sure things changed with the 386 being 32 bit and having a much better memory manager setup but the 286 platform wasn't bad for its time.
 
The 286-20/25 would have been faster clock per clock then the 386SX and motherboards that supported those speeds probably used normal SIMM sockets so memory expansion was cheap and easy.

Just out of curiosity, under what circumstances would the 286 be faster *per clock* than the 386sx? I mean, I guess an edge case you might be able to point to would be if the 386sx is running a memory manager that puts it in VM8086 mode, that might add a little latency compared to a 286 that has full hardware EMS support built into the motherboard chipset, but outside of that?

There was this brief moment where 20 and 25mhz 80286s looked like a hot deal for running DOS software because you could get one for three or four hundred bucks less than a 16mhz 386sx which, yeah, they’d totally run faster than, but it was the clock speed difference responsible for that. It was all about price, though. And I wouldn’t necessarily say that people who chose the fast 286s were necessary wrong at the time because if you were happy with single-tasking DOS programs the premium Intel was milking out of the SX was pretty steep. (I upgraded my dog-slow 4.77mhz XT clone into an ad-hoc 12mhz 5162 wannabe sometime in 1989 or so because you could get the 286 motherboard for peanuts… $100-or so bucks?, while effectively the same board with a 386sx on it was around $500, minimum.

Windows 3.0 was undoubtedly a major factor in killing the 286, but I think what really finished it off was AMD cloning the 386 and forcing prices to find their natural level. Without the monopoly premium on the silicon there’s zero reason to prefer the 286.
 
I had never given it that much thought. I had always considered the '286 machines to be just another beige box in the 40-year procession of them. It so happens my first "PC" was a 286 clone which I bought from one of those tine mom-and-pop operations that advertised in the Bay Area classifieds. It had perhaps 1M, maybe 2M of RAM and a 10 or 20 meg disk. It ran DOS (fortified by 4DOS and a bunch of utilities from SimTel), and Coherent, until I was compelled to buy a 386sx at the beginning of December 91 so I could run Linux (boot and root, no login). I liked the 286 well enough, and the 386 was nice (and had 4MB RAM), but I had no sentiment for either machine like I had for the cranky OSI I had as a kid.

Objectively, I know that these machines are outside my "imprinting" window, but I still have a hard time getting my head around the fact that some folks are sentimental for these machines, which I always regarded as lacking in personality.

Still, given the numbers of these machines, I am surprised it's so hard to find one.
 
The realistic possibility I could imagine software with different optimization paths for "below 386" and "386+", where the "386+" path assumed "386DX+" and used 32-bit memory accesses gratuitiously. On a narrow-bus SX system, that could be a big hit.

I could imagine chipset/mainboard design optimization being a thing, but a 386SX is close enough electrically that if you made a world-beating 286 motherboard, it would make sense to tweak it to package it as a 386SX as well/instead.

More practically, I could see it as a full-system thing. The 286/20 at a given price point might have more memory, faster memory/fewer wait states, a faster drive, or a better graphics card than a 386SX/20.

I feel like in a way, the PC world's backwards-compatibility obsession prevented the 286 from thriving. The people who bought them as basically a faster XT never meant much pressure on software designers to exploit the processor's features. By the time lots of people were grumbling about the 640k limit, they could just target the 386.

If the AT had been a more blatant break with compatibility-- say, if it was MCA, or if its BIOS left you in protected mode at the end of boot-- the people who bought them would have likely been expecting the full 286 experience. They'd probably mostly ship with Unix-flavoured operating systems, or we'd see some much earlier forced attempts to get a protected-mode DOS-style product out the door.

On the other hand, it's entirely possible by the time the AT hit the market, the PC/XT juggernaut was already unstoppable. In that scenario, the incompatible-AT ends up a niche product, and we'd have some non-IBM vendor eventually bolt a 286 into a 5160-compatible design but without necessarily the same market-shaping force to create a universal standard. I'm sort of thinking of something like the Olivetti M24/AT&T 6300, which already had a "sort of 16-bit" expansion slot.

Given the long-lived market for "just a faster XT", I'm surprised that most 8086-class machines top out at 10 or maybe 12MHz. You could reasonably use faster embedded 80186/188/V40/V50 parts, and ignore their onboard peripherals to get even more mileage out of that architecture. It would be hilarious to see a modern 50MHz 80188 shoehorned into a conventional 8-bit ISA system, likely with the internal peripherals deactivated and a shedload of wait-states.
 
The Feb 8, 1988 Infoworld indicated that the 286 and 8086 were selling at about the same rate in 1987 with maybe 10% of the market going to the 386. By 1990, the 8086 had disappeared from the desktop market. Using roughly equivalent systems, the 286-20 was $300 cheaper than the 386SX-20 and $600 cheaper than the 386-20. Given the 1990 floppy only 286-16 AT clone cost $600, there wasn't much room to have a cheaper 8088/8086 design and I doubt anyone would have sprung for the cache needed to support a very high speed 8086. I did buy a 286 clone back in 1987; strangely, it was only about $50 more than the same companies XT-clone with an identical hard disk.

The 286 had a good run despite the pressure of better CPUs dropping in price. The 286, unfortunately, had its market expansion hampered by the failure of the 16-bit protected mode OSes like OS/2 1.x and Xenix 286.
 
Just out of curiosity, under what circumstances would the 286 be faster *per clock* than the 386sx? I mean, I guess an edge case you might be able to point to would be if the 386sx is running a memory manager that puts it in VM8086 mode, that might add a little latency compared to a 286 that has full hardware EMS support built into the motherboard chipset, but outside of that?

There was this brief moment where 20 and 25mhz 80286s looked like a hot deal for running DOS software because you could get one for three or four hundred bucks less than a 16mhz 386sx which, yeah, they’d totally run faster than, but it was the clock speed difference responsible for that. It was all about price, though. And I wouldn’t necessarily say that people who chose the fast 286s were necessary wrong at the time because if you were happy with single-tasking DOS programs the premium Intel was milking out of the SX was pretty steep. (I upgraded my dog-slow 4.77mhz XT clone into an ad-hoc 12mhz 5162 wannabe sometime in 1989 or so because you could get the 286 motherboard for peanuts… $100-or so bucks?, while effectively the same board with a 386sx on it was around $500, minimum.

Windows 3.0 was undoubtedly a major factor in killing the 286, but I think what really finished it off was AMD cloning the 386 and forcing prices to find their natural level. Without the monopoly premium on the silicon there’s zero reason to prefer the 286.

I've benchmarked 286-16 vs 386sx-16 on identical setups. Same chipset, zero wait, no EMM. The 286 was consistently about 5-10% faster
 
The realistic possibility I could imagine software with different optimization paths for "below 386" and "386+", where the "386+" path assumed "386DX+" and used 32-bit memory accesses gratuitiously. On a narrow-bus SX system, that could be a big hit.

Are you suggesting that there's code that will run faster on an SX than on a DX? That code "written for the DX" would be "extra slow" on the SX than code written specifically for the SX?
 
Back
Top