Benefits of building a 386 computer

In short... It's like for building any other old computer... nostalgia reasons.

kixs wrote on 2024-12-13, 08:18:

Benefits of building a 386 computer

In short... It's like for building any other old computer... nostalgia reasons.

And history reasons. And compatibility reasons. :)

Personally, I would set up a 386 in PCem/86Box first and then decide if it suits my demands.
It's just an approximation, but it gives a good impression I think.

What I like about 386 PCs has been already mentioned, I think, so I'll try to hold back now.

I just think it's more complex situation, because the 386 wasn't a random chip. It was as influenzal as the i4004 or the Z80.
The 386 was the first 32-Bit x86 CPU and once dubbed "mainframe on a chip".

Edit: "[..] past the 8086 that made the IBM PC run, through the 80386 that they called a "mainframe on a chip" [..]"
Source: https://www.cuug.ab.ca/branderr/pmc/010_moore … _law_intel.html

Some say the 386 design predates the 286, but the the 286 was fabricated first due to limits of the technology of the day.

Edit: This one is good.
"You might think of the Intel 386 processor (1985) as just an early processor in the x86 line, but the 386 was a critical turning point for modern computing in several ways.
1 First, the 386 moved the x86 architecture to 32 bits, defining the dominant computing architecture for the rest of the 20th century.
The 386 also established the overwhelming importance of x86, not just for Intel, but for the entire computer industry.
Finally, the 386 ended IBM's control over the PC market, turning Compaq into the architectural leader."
http://www.righto.com/2023/10/intel-386-die-v … rsions.html?m=1

Then there's also smoothness. Yes, you can play Commander Keen on a Pentium IV @3,8 GHz with a 512 MB GPU.
But is it as smooth as on a 386 with an ISA VGA card? (Or EGA card)

The problem with fast PCs that are being underclocked/slowed down is that the process isn't happening evenly.
If you disable caches and reduce FSB, then the efficient pipelineing and one-cycle instructuons of the modern CPU doesn't automagically go away.

Depending on how the game was coded, there might be micro-stutter, scrolling issues and so on.

‐
Edit: -> Btw, there's something you should know about games.
Historically, the IBM PC had been quite good at 3D games such as MS Flight Simulator, but not 2D games!

So running Quake, Duke Nukem 3D, Wolf 3D or Doom smoothly on you Celeron-based DOS PC is no indication for fine 2D gaming, as well.

That's because the PC graphics hardware can't do scrolling; well at least not without VGA CRTC magic or using EGA's dual-page mode.
- Which in turn means that the VGA card must be of very good quality/very good compatibility to make the magic work.

It's funny to think that an old NES or Sega Genesis/MD had better 2D capabilities than your average 486 PC.
Especially the Sega Genesis/MD was notable for having good parallax scrolling.
-

Last but not least, a lot of the legacy blocks in modern PCs are just there for backwards compatibility. On a superficial level. No efforts are being made keep things faithfully.

VGA compatibility in PCI/AGP/PCIe graphics cards, for example, is just there to allow Windows Setup to be run once. No one cares if colours palettes are wrong or if there are artifacts. It's just a tool.
It's not there to run demoscene productions or there to be hacked to generate non-standard TV signals (like French 819 line TV).

On older, ISA, EISA or VLB VGA cards this was different. DOS and industrial operating systems like Real/32 or *nix systems were still "living".
The industry needed good hardware compatibilty, because non-consumer operating systems did rely on generic device drivers.

With the success of Windows 95 this had changed.
Hardware release afterwards didn’t really bother with hardware compatibilty anymore.

Lower end sound cards (on-board, AC97 etc) started to emulate Sound Blaster through VXDs rather than having SB/SB Pro compatibilty in silicon.

Because, DOS games are to be run from Windows anyway, right? It's the future! No one needs plain DOS if there's Windows! Or so the industry thought.

Another reason for having a 286/386/486 is that certain games did support specific hardware of a given era.
Like the Action Replay ISA card for DOS games, ET-4000AX and Paradise for SVGA modes in games, the CH yoke for flight sims, timing issues with joysticks, speed-sensitive AdLib games etc.

Same goes for the 486 era, maybe.
Early virtual reality systems may require a fast 486 system in order to run properly (486DX2-66 minimum), along with proprietary graphics cards for ISA or VLB slot.

Not sure if that's about nostalgia or just practibility, though.

I mean, same could be said about relevance of an Amiga.
Who needs an Amiga if there's an Atari ST? Games are same on both platforms. ;)

Edit: I read there were some games that made use of the undocumented op-codes of the 486.
They thus only ran on 486DX and 486DX2 (original i486 design), but not the 486DX4 and later CPUs.
Patches had to be used to get them running.

Edited.

Jo22 wrote on 2024-12-13, 15:51:

The problem with fast PCs that are being underclocked/slowed down is that the process isn't happening evenly.
If you disable caches and reduce FSB, then the efficient pipelineing and one-cycle instructuons of the modern CPU doesn't automagically go away.

Depending on how the game was coded, there might be micro-stutter, scrolling issues and so on.

Very true. I think some games, like Wing Commander 1 are tricky on faster hardware, as the speed varies by the number of enemies around. So it may be ok in parts, then too fast later as you clear the enemies.

.. what's also cool about early 80386 systems is their sheer size. 😁
Here's an board that measures 15.75 by 20 inches. That's 40 by 50,8 cm! It has an 80287 socket, I think.
https://x.com/TubeTimeUS/status/1535417062553554945

The greatest appeal for a 286 to me is the "wow, this actually works?!" factor, especially with a hotrod 25Mhz 0WS system - it's genuinely surprising just what can be pulled off with experimentation and a bit of determination and is some of the most fun I have with retro PC tinkering. I've spent more time discovering the hidden gem of 286 possibilities more than anything else in this hobby - I could throw away 90% of hardware I have that can't/doesn't work on a 286 and still be largely content as the discovery factor and the generally dismissive attitude towards 286s that can be seen around here are good personal motivators. It's significantly less painful than having Win98SE self-destruct multiple times in a single day, like I had to deal with recently 😀

Finding commercial DOS games as late as 1995 that still run at a playable speed on a 286 (and even run on an 8088, if you're really determined) is a marvel by itself - the lack of a external 32-bit extender like DOS4GW/RTM/CWSDPMI0.EXE and lack of strings that can be found easily enough in game EXEs (ie. "P/MODEW", "DOS/4G", DJGPP headers which immediately stand out for being several lines long, some variant of "80386 processor required!" etc) are good targets to try - sometimes you'll come across a game that explicitly requires 386 instructions without any indication, but there are rare instances where if a game has separate intro/game executables (two examples that immediately come to mind are Entity, and Machines that was published by Merit Studios in 1995) you can simply bypass the intro EXE if it requires 386 instructions. Something I've learned is that the minimum requirements listed on boxes like to lie more than you think. Plenty of emulators for 8-bit systems also work fine, while some ZX Spectrum/ZX81/VIC-20/TRS80 emulators can hit full speed without issues and the nocash Amstrad/MSX/GB emulators can hit playable speeds, others are more for the cool factor more than practicality as there isn't simply enough juice to get full speeds (ie. NESA for NES, XFormer 2.52 for the Atari 800/XL, C64s 0.9a (maybe 0.9b? haven't confirmed) for C64).

Want a good accelerated Windows 3.1 experience? Sure thing, grab yourself a ATI mach8/mach32 or a S3 911/924 and go to town, just make sure you find the right drivers. WD90C31 cards are good if you don't care about font rendering acceleration or non-orthagonal line drawing, all Cirrus Logic GD54xx cards (except maybe the GD5401) are surprisingly competent as well and perform much better with font rendering and some blitter operations than framebuffers, and certainly beats the WD90C31 with font rendering. The Avance Logic ALG2101 is a rare card to find, but is a step up from the WD90c31 in the sense that it does accelerate non-orthagonal lines, though it's best at 640x480x256 - it uniquely loses performance with higher resolutions. The best card for a combined DOS/Win 3.1 experience is a VRAM mach32 - DOS speeds are at worst ~3% slower than top speed ISA cards, outside of Wolf3d/Doom8088/MicroProse F1GP/Eye of the Storm you probably won't even notice the difference. WD90c3x cards do have palette cycling issues in DOS with fade outs, but are relatively minor compared to a few other cards I've used.

Sadly, 286/Standard Mode Windows 3.1 games/apps is very poorly documented in general and not really centralised - I want to fix that in due time, and have been slowly working on that behind the scenes. More posts like Video playback on a 286? - here's how to do it and Re: Windows for Workgroups TCP/IP Disk are sorely needed.

I haven't even looked into ELKS yet, but that intrigues me a lot.

Like everyone else, there's a nostalgia factor attached to why I care about 286s - even though the first PC my family had after the Amiga 500/1200 was a 486, the first time we have two PCs at the same time was a hand-me-down 286/16 with 2MB of RAM, VGA card and a sound card (though it probably was only PC speaker output with external speakers that had volume control, memory is a bit fuzzy there) from one of my great-uncles, and I was the one who used it 90% of the time since I was the youngest - I didn't care though, Wolf3D/Spear of Destiny, Stunts, Tunnels of Armageddon, Jones in the Fast Lane, Captain Comic, Keen 4 and Bio Menace all kept me occupied. I still have a few Stunts levels I made on that 286.

A 386SX-16 interests me just to see how the absolute rock bottom of 386 processors can handle 32-bit games/apps - same thing with the 486SLC for things that require 486 instructions.

The SX16 gives me much the same feeling as 6Mhz 5170s, 1.3 or 1.4 P4s, the "Ick, why does this exist?" from lack of "snap" and general feel of comparative willingness compared to the better CPU of previous gen.

I am currently working on a Packard Bell 286, trying to track a parity fault. I may have misjudged that in the past, last ran it 2 decades back and thought it was a doggy 10 or 12 Mhz. I had not realised then that this machine,should be 16Mhz, and may have three turbo levels, so I might not have got it out of second gear or something. Also this NEAT chipset is supposed to do 4 way interleave "virtually zero wait state" (I think that's the reason for zero wait state claims from that era, not abundantly available 50ns or faster RAM ) So set up right, it should be quite a snappy little fella. I mean I've been spoiled having run a 25mhz Harris board back in the day and finding it was "too fast" for earlier versions of Wolf3D (1.4 and prior I think) but 16mhz should sit nicely between my fastest XT and "reasonable" 386es, like 25mhz.

My Headland HT12 mobo has two oscillator clocks for controlling the CPU and a jumper to switch between the two - it originally had a 24Mhz and 32Mhz clock that were both soldered, but I replaced the 24Mhz with a 50Mhz clock and socket to get the most out of it - if I'm feeling particularly nostalgic, I can always swap the jumper position and disable half of the RAM and more or less have the same experience I had back in the mid-90s. Can't say I recall what video card was installed back then but I didn't need to decrease the window size in Wolf3D/SoD too much to get a playable experience, two or three notches at best so I'm willing to make a safe bet that the mobo was set to 0WS. Wouldn't be surprised if it was a Trident 8900CL/D, WD90C30 or a Cirrus Logic GD5402.

DEAT wrote on 2024-12-17, 03:27:

The greatest appeal for a 286 to me is the "wow, this actually works?!" factor, especially with a hotrod 25Mhz 0WS system - it's genuinely surprising just what can be pulled off with experimentation and a bit of determination and is some of the most fun I have with retro PC tinkering. I've spent more time discovering the hidden gem of 286 possibilities more than anything else in this hobby - I could throw away 90% of hardware I have that can't/doesn't work on a 286 and still be largely content as the discovery factor and the generally dismissive attitude towards 286s that can be seen around here are good personal motivators. It's significantly less painful than having Win98SE self-destruct multiple times in a single day, like I had to deal with recently 😀

...

A 386SX-16 interests me just to see how the absolute rock bottom of 386 processors can handle 32-bit games/apps - same thing with the 486SLC for things that require 486 instructions.

Someone is working on a new Super-286, and is considering the potential of forward compatibility to 386SX, 486 etc.
https://forum.vcfed.org/index.php?threads/pro … m-5170.1243108/

For me, I think someone who has spent some time using these types of systems in the early 90s, can understand they are unique. They really kind of were made for the time when DOS was king, even if they aren't the fastest systems for DOS. They could be used like the first PC, and run DOS, but they can also run more types of OS, and have more memory if you want, even when people back then didn't really need to.

I think I would have liked one of those compact baby 386 motherboards. Maybe if I had one, I would have kept one all this time. I tossed everything, 286, 386, 486, I had for space reasons, back in the late 90s. But maybe will get one again. I'd be happy with a 286, most makes, but 386SX or 486SLC, all I've used at one point would be cool too, just like you mention. I'd even take a 386DX (also another one I had), but those are getting a bit high end 😉

DEAT wrote on 2024-12-17, 03:27:

The greatest appeal for a 286 to me is the "wow, this actually works?!" factor

Same here, but for slightly different reason. In my opinion 286 was the first almost usable x86 home PC around 1990.
- Huge amount of games, ~1900 games between 1989-1991. Second popular was C64 with ~1400 game in same period.
- Most games supported Adlib. Still barely any Sound Blaster support tho First dos sound blaster game. Few 1989 game, tens of games in 1990, up to ~200 by 1991.
- VGA became mainstream pushing EGA out.
- Windows finally started making sense with release of 3 in 1990.

1990 was the year Amiga home superiority started showing slight cracks. Games were coming out for both platforms, but fast 286 with VGA often delivered better visuals and smoother gameplay in 3D titles at the 2-3x $$$ premium. As I wrote in Re: Ideal gaming system specs per year? (Esp. 1983 - 1992?)

1990: 286 12-16MHz / 1MB ram / VGA finally got cheap enough for home / adlib

1991: Little more mhz and maybe 386sx / maybe 1MB more ram / VGA / adlib or maybe Sound Blaster
Very little SB support. Not much using more than 1MB ram. Not much needing faster CPU.

1992: Flood of cheap Am386DX, 486SX / 4 MB ram / VGA / Sound Blaster
Wolf 3d, Ultima Underworld, Dune 2, Stunt Island, Alone in the Dark

1992 is where PC really started shining thanks to cheap 386 machines, it was the beginning of PC master race era. Of course Im also biased because my first PC was Am386DX 40MHz / 8MB / VGA / SB / CD 😀

The greatest appeal for a 286 to me is the "wow, this actually works?!" factor

the actual wow factor here is that this processor was released in 1982 (!), so about the same time as the c64. The C64 STARTED strong but the 286 ENDED strong...😀

DEAT wrote on 2024-12-17, 03:27:

Finding commercial DOS games as late as 1995 that still run at a playable speed on a 286 (and even run on an 8088, if you're really determined) is a marvel by itself - the lack of a external 32-bit extender like DOS4GW/RTM/CWSDPMI0.EXE and lack of strings that can be found easily enough in game EXEs (ie. "P/MODEW", "DOS/4G", DJGPP headers which immediately stand out for being several lines long, some variant of "80386 processor required!" etc) are good targets to try - sometimes you'll come across a game that explicitly requires 386 instructions without any indication, but there are rare instances where if a game has separate intro/game executables (two examples that immediately come to mind are Entity, and Machines that was published by Merit Studios in 1995) you can simply bypass the intro EXE if it requires 386 instructions. Something I've learned is that the minimum requirements listed on boxes like to lie more than you think. Plenty of emulators for 8-bit systems also work fine, while some ZX Spectrum/ZX81/VIC-20/TRS80 emulators can hit full speed without issues and the nocash Amstrad/MSX/GB emulators can hit playable speeds, others are more for the cool factor more than practicality as there isn't simply enough juice to get full speeds (ie. NESA for NES, XFormer 2.52 for the Atari 800/XL, C64s 0.9a (maybe 0.9b? haven't confirmed) for C64).

I was very surprised a couple of weeks ago when I found out Apogee's Realms of Chaos from 1995 is a real mode game. It uses 386 instructions, but it does run on a 286 when EMU386 is used.

Jo22 wrote on 2024-12-13, 15:51:

I just think it's more complex situation, because the 386 wasn't a random chip. It was as influenzal as the i4004 or the Z80. Th […]
Show full quote

I just think it's more complex situation, because the 386 wasn't a random chip. It was as influenzal as the i4004 or the Z80.
The 386 was the first 32-Bit x86 CPU and once dubbed "mainframe on a chip".

Edit: "[..] past the 8086 that made the IBM PC run, through the 80386 that they called a "mainframe on a chip" [..]"
Source: https://www.cuug.ab.ca/branderr/pmc/010_moore … _law_intel.html

In my Domesticating the Computer VCF talk, at the very end I had this chart I was putting together.

The main point being just to show the 386 was the point when hobbyist microcomputers started to match the processing capability of existing minicomputers. To me, this makes the 386 more historically significant than say the 486 or Pentium.

Prior to that point (in about 1986): microcomputers were mostly viewed as fancy multi-line calculators, while minicomputers still cost more than a house (you could get 64KB minicomputers with a keyboard and serial port as early as 1968, even FORTRAN, BASIC, or APL "dev environment", just they were over $30K and still used a line printer for output).

Then as others mentioned, the VGA standard plus the CompuServe GIF image format standard happening in 1987. Multimedia really starting to become a thing then.

So I can agree with what was said earlier, like that Sept 1986 InfoWorld article represents: they knew the 386-based system was a pivotal change to the industry, and (to me) marks the end and decline of the TVT-based character-generator and boot to BASIC systems (which were all formally abandoned not until 1993).

Or said another way, the idea of "a computer in every home" didn't really happen until sometime after the 386 release. Not just based on the cost of the systems, but on them having enough utility to make it worthwhile to home users. Here is another chart I didn't get to include at the time:

- the big jump in Apple2 sales in 1979 was largely attributed to VisiCalc being first released on that system (then soon afterwards on PET then other systems)
- The next big jump in Apple2 sales is said to be attributed to the Z80/CPM card (1981)
- the IBM PC struggles at first, but by 1986 it basically becomes "off the charts" relative to everything-else-combined
- It's not talked about a lot, but the Altair had a somewhat sad demise (just quickly supplanted by IMSAI and other systems)
- the huge leap in C64 and VIC-20 in 1982/1983 is said to be attributed to that offer posted in the New York Times

voidstar wrote on 2025-05-23, 07:50:

In my Domesticating the Computer VCF talk, at the very end I had this chart I was putting together. […]
Show full quote

Jo22 wrote on 2024-12-13, 15:51:

I just think it's more complex situation, because the 386 wasn't a random chip. It was as influenzal as the i4004 or the Z80. Th […]
Show full quote

I just think it's more complex situation, because the 386 wasn't a random chip. It was as influenzal as the i4004 or the Z80.
The 386 was the first 32-Bit x86 CPU and once dubbed "mainframe on a chip".

Edit: "[..] past the 8086 that made the IBM PC run, through the 80386 that they called a "mainframe on a chip" [..]"
Source: https://www.cuug.ab.ca/branderr/pmc/010_moore … _law_intel.html

In my Domesticating the Computer VCF talk, at the very end I had this chart I was putting together.

The attachment minimicroconvergence.jpg is no longer available

The main point being just to show the 386 was the point when hobbyist microcomputers started to match the processing capability of existing minicomputers. To me, this makes the 386 more historically significant than say the 486 or Pentium.

Prior to that point (in about 1986): microcomputers were mostly viewed as fancy multi-line calculators, while minicomputers still cost more than a house (you could get 64KB minicomputers with a keyboard and serial port as early as 1968, even FORTRAN, BASIC, or APL "dev environment", just they were over $30K and still used a line printer for output).

Then as others mentioned, the VGA standard plus the CompuServe GIF image format standard happening in 1987. Multimedia really starting to become a thing then.

So I can agree with what was said earlier, like that Sept 1986 InfoWorld article represents: they knew the 386-based system was a pivotal change to the industry, and (to me) marks the end and decline of the TVT-based character-generator and boot to BASIC systems (which were all formally abandoned not until 1993).

Or said another way, the idea of "a computer in every home" didn't really happen until sometime after the 386 release. Not just based on the cost of the systems, but on them having enough utility to make it worthwhile to home users. Here is another chart I didn't get to include at the time:

- the big jump in Apple2 sales in 1979 was largely attributed to VisiCalc being first released on that system (then soon afterwards on PET then other systems)
- The next big jump in Apple2 sales is said to be attributed to the Z80/CPM card (1981)
- the IBM PC struggles at first, but by 1986 it basically becomes "off the charts" relative to everything-else-combined
- It's not talked about a lot, but the Altair had a somewhat sad demise (just quickly supplanted by IMSAI and other systems)
- the huge leap in C64 and VIC-20 in 1982/1983 is said to be attributed to that offer posted in the New York Times

The attachment initialminis.jpg is no longer available

I agree that 80386 and the architectural improvements it introduced over the predecessors make it one of the most influential x86 CPUs of all time. It is certainly much more important than 486 which more or less just built on what was created with 386. However, I’d generally argue that P5 is pretty on par at least. Although it has largely the i386 instruction set that was already in 386 and 486, it introduced new features that were before seen in big iron systems or outside of the x86 ecosystem often on much more expensive platforms. It was the first superscalar x86, first x86 with proper multi-cpu support, 64-bit data bus, separate data and instruction caches etc. P5 really brought x86 on par with many newer architectures, such as Sun SPARC, DEC Alpha, PowerPC 601 etc. Without P5 ”leap” I think it is very possible that x86 would’ve faced much more serious competition during the 90s and Intel could’ve found itself in similar situation that Motorola was facing with their aging 68k design.

i386 was what brought x86 on par with 68K and P5 was a colossal leap in home computing power that would more or less be the defining architecture all the way up until AMD introduced x86-64. And that is my point here, I think i386, iP5 and Amx86-64 are the three milestone architectures for x86..

^No way! The 286, 386's little brother, was quicker than 68000 at same clock speed! :D
Edit: Never mind. I just couldn't resist! ;)

I noticed we are being a bit vague and general, "improvements" "extensions" the big thing, the elephant we are avoiding is...

MMU

The 386 (and 68030) onboard MMU allowing virtual memory in gigabytes of address space, proper protected address spaces and thus virtualization facilities. MMU stands for Memory Management Unit, and that's what it did, no single user, single task, if you bump into your own data it's your fault, stuff here.

This is what made 386 a mainframe/minicomputer on a chip. That's what made Torvalds go "Hey, I've got a minicomputer on a chip, there oughta be some sort of advanced OS for that...."

You could almost surmise that DEC had been Intel's best salesperson, hyping up and "pre-running" the market for virtual memory features with the VAX range, several years in advance, but getting it into the heads of IT consumers, "This is the real workhorse, this is what you want." a decade in advance of it coming to the desktop.

BitWrangler wrote on 2025-05-28, 13:18:

I noticed we are being a bit vague and general, "improvements" "extensions" the big thing, the elephant we are avoiding is... […]
Show full quote

I noticed we are being a bit vague and general, "improvements" "extensions" the big thing, the elephant we are avoiding is...

MMU

The 386 (and 68030) onboard MMU allowing virtual memory in gigabytes of address space, proper protected address spaces and thus virtualization facilities. MMU stands for Memory Management Unit, and that's what it did, no single user, single task, if you bump into your own data it's your fault, stuff here.

This is what made 386 a mainframe/minicomputer on a chip. That's what made Torvalds go "Hey, I've got a minicomputer on a chip, there oughta be some sort of advanced OS for that...."

You could almost surmise that DEC had been Intel's best salesperson, hyping up and "pre-running" the market for virtual memory features with the VAX range, several years in advance, but getting it into the heads of IT consumers, "This is the real workhorse, this is what you want." a decade in advance of it coming to the desktop.

The MMU was just a necessary stepping stone to allow the 386 to do virtual 8086 mode. Virtual 8086 mode is why the 386 thrived but the 68030 didn't, IMHO. a 386 could do a better DOS than DOS, whether that was Loading your drivers into upper memory, QEMM, Windows 3.0 Enhanced Mode, or OS/2 2.0. I did all of those things on my 20Mhz 386sx.

BitWrangler wrote on 2025-05-28, 13:18:

I noticed we are being a bit vague and general, "improvements" "extensions" the big thing, the elephant we are avoiding is...

MMU

Exactly this.

LaurieWired touches on this in this video: https://www.youtube.com/watch?v=vc79sJ9VOqk

The idea of an MMU was around since the 1960s, and incorporating that into a microprocessor chip is when personal desktop systems finally matched the capability of prior-decade mainframes. Prior to this, micros weren't much more than fancy desktop calculators.

Even Bill Gates knew the 286 was a stepping stone to what was truly needed. It is rumored that part of why DR (digital research) dragged their feet on CP/M for the IBM PC, is he was trying to complete a concurrent/multitasking version of an x86 DOS. And there are 8086 based multitasking systems, they just don't work very well (GRiD-OS kind of worked ok an 8086). But with a 386, then something like OS/2 2.0 becomes possible.

But in the 1960s - mainframes were so expensive to acquire and operate, an "idle CPU" was unthinkable. And so basically from Day 1, they were designed for multi-user and time-sharing - if that mainframe is running, it better be calculating something useful for somebody. That's why the idea of a "personal computer" was so laughable in the early 1970s - no individual could afford the per-hour operating cost. But in addition to the microprocessors, I think we often overlook the improvement in power supply quality, and how quickly CRTs came down in price right around 1971. That said, the earliest microprocessors weren't very good - you could darn near calculate by hand faster than those (which is why systems then still used regular TTL, like in the Datapoint and Alto). Sure, the big companies still wanted to just lease equipment - so the computer liberation movement (c. 1973/1974) was quite justified. And while 1977 was a pivotal year for micros (they were finally capable enough to run a BASIC parser), one argues they didn't really become useful systems until at least a few years later.

And there are 8086 based multitasking systems, they just don't work very well (GRiD-OS kind of worked ok an 8086). But with a 386, then something like OS/2 2.0 becomes possible.

Depends. Generally true, I think.

Concurrent DOS 286, Xenix 286 and Unix System V did run reasonable stable on an 80286.

Some had the ability to multitask DOS applications, even.
Like Concurrent DOS 286 (later renamed CCDOS XM) and Unix System V (solely software-based via Merge 286).

Or PC-MOS/386 and Wendin DOS, on plain 8086/8088 and up.
V86 wasn't needed here, though memory was limitting factor in Real-Mode.

I noticed we are being a bit vague and general, "improvements" "extensions" the big thing, the elephant we are avoiding is...

MMU

The 80286 introduced the MMU on x86.
80386 added paging unit and segments bigger than 64KB; prior the 286 already could go smaller than 64KB.
But there's more. There were external MMUs like the ALLCHARGE card or SodaPop card.
https://www.youtube.com/watch?v=xgMIbo6QoM4

These were 286 daughter boards with an programmable MMU, which also could support EMS.
PC-MOS/386 supported them, I think.

Then there were hardware add-ons like the Merge hardware.
It allowed trapping, so that the software could catch misbehaved DOS software.
https://en.wikipedia.org/wiki/Merge_(software)

On prior x86 hardware, there were discreet MMUs.
Siemens PC-X (80186) had an MMU to run Sinix, an 16-Bit version of Unix.
The version stripped from MMU was sold as PC-D for running MS-DOS.
https://en.wikipedia.org/wiki/Siemens_PC-D

Such MMU features found their way into 286 chipsets and x86 SoCs, in the form of EMS hardware.
Some also provided virtual memory or a bigger address space.

The newer NEC V-Series processors, for example, had 286-ish features without Protected-Mode.
Then there was the F8680 integrated PC/XT hardware, with 26-Bit (64MB) address bus.
https://datasheets.chipdb.org/Chips/F8680_rev_1_1.pdf

PS: Then there's one detail that historians always seem to either ignore or forget, when talking about DOS compatibility and V86: The OS/2 Family API (FAPI)

Because if everything had went as planned by IBM, then multitasking DOS applications on 80286 would have been a non-issue.

Long story short, Family API applications are OS/2 applications that can run on DOS.
In the header, there's a basic OS/2 runtime (MZ stub) that executes if the application is run from ordinary DOS.

Such FAPI applications are native OS/2 applications (16-Bit NE) that can run on both DOS and OS/2 in short.
But on OS/2, they can make use of megabytes of memory, memory protection and multitasking.

And no, it's not some weird niche stuff that never "took off".
By late 80s, many major compilers had supported FAPI, such as Microsoft Basic PDS 7.1 and MASM.
Even Windows NT could run these "DOS" applications natively, until the 16-Bit OS/2 sub system got canned. OS/2 Warp also supported them.

More information here (not affiliated):
https://www.os2museum.com/wp/os2-history/os2-1-0/
http://www.malsmith.net/blog/visual-c-os2-family/
http://www.os2voice.org/vnl/past_issues/vnl0801h/vnewsf4.htm