When Windows 2 was launched, there were two editions: Windows, and Windows/386.
The ordinary "base" edition of Windows 2.0x ran on an XT-class computer: that is, an Intel 8088 or 8086 CPU. These chips can only directly access a total of 1MB of memory, of which the highest 384kB was reserved for ROM and I/O: so, a maximum 640kB of RAM. That was not a lot for Windows, even then. But both DOS and Windows 2.x did support expanded memory (Lotus-Intel-Microsoft-specification EMS). I ran Windows 2 on 286s and 386s at work, and on 386 machines I used Quarterdeck's QEMM386 to turn the extended memory that Windows 2 couldn't see or use into expanded memory that it could.
The Intel 80286 could access up to 16MB of memory. But all except the first 640kB was basically invisible to DOS and DOS apps. Only native 16-bit programs could access it, and there barely were any — Lotus 1-2-3 r3 was one of the few, for instance.
There was one exception to this: due to a bug the first 64kB of memory above 1MB (less 16 bytes) could be accessed in DOS's Real Mode. This was called the High Memory Area (HMA). 64kB wasn't much even then, but still, it added 10% to the amount of usable memory on a 286. DOS 3 couldn't do anything with this – but Windows 2 could.
Windows 2 and 2.01 were not successful, but some companies did release applications for them – notably, Aldus' PageMaker desktop publishing (DTP) program. So, Microsoft put out some bug-fix releases: I've found traces of 2.01, 2.03, 2.11 and finally 2.12.
When Windows 2.1x was released, MICROS~1 did a little re-branding. The "base" edition of Windows 2.1 was renamed Windows/286. In some places, Microsoft itself claims that this was a special 286 edition of Windows 2 that ran in native 80286 mode and could access all 16MB of memory.
But some extra digging by people including Mal Smith has uncovered evidence that Windows/286 wasn't all it was cracked up to be. For one thing, without the HIMEM.SYS driver, it runs perfectly well on 8088/8086 PCs – it just can't access the 64kB HMA. Microsoft long ago erased the comments to Raymond Chen's blog post, but they are on the Wayback Machine.
So the truth seems to be that Windows/286 didn't really have what would later be called Standard Mode and didn't really run in the 286's protected mode. It just used the HMA for a little extra storage space, giving more room in conventional memory for the Windows real-mode kernel and apps.
So, what about Windows/386?
The new 80386 chip had an additional mode on top of 8/16-bit (8088/8086-compatible) and fully-16-bit (80286-compatible) modes. The '386 had a new 32-bit mode – now called x86-32 – which could access a vast 4GB of memory. (In 1985 or so, that would have cost hundreds of thousands of dollars, maybe even $millions.)
However, this was useless to DOS and DOS apps, which could still only access 640kB (plus EMS, of course).
But Intel learned from the mistake of the 286 design. The 286 needed new OSes to access all of its memory, and even they couldn't give DOS apps access to that RAM.
The 386 "fixed" this. It could emulate, in hardware, multiple 8086 chips at once and even multitask them. Each got its own 640kB of RAM. So if you had 4MB of RAM, you could run 6 separate full-sized DOS sessions and still have 0.4MB left over for a multitasking OS to manage them. DOS alone couldn't do this!
There were several replacement OSes to allow this. At least one of them is now FOSS -- it's called PC-MOS 386.
Most of these 386 DOS-compatible OSes were multiuser OSes — the idea was you could plug some dumb terminals into the RS-232 ports on the back of a 386 PC and users could run text-only DOS apps on the terminals.
But some were aimed at power users, who had a powerful 386 PC to themselves and wanted multitasking while keeping their existing DOS apps.
My personal favourite was Quarterdeck DESQview. It worked with the QEMM386 memory manager and let you multitask multiple DOS apps, side by side, either full-screen or in resizable windows. It ran on top of ordinary MS-DOS.
Microsoft knew that other companies were making money off this fairly small market for multitasking extensions to DOS. So, it made a
Windows programs, including the Windows kernel itself, still ran in 8086-compatible Real Mode and couldn't use all this extra memory, even on Windows/386. All Windows/386 did was provide a loader that converted all the extra memory above 1MB in your high-end 386 PC – that is, extended (XMS) memory – into expanded (EMS) memory that both Windows and DOS programs could use.
The proof of this is that it's possible to launch Windows/386 on an 8086 computer, if you bypass the special loader. Later on, this loader became the basis of the EMM386 driver in MS-DOS 4, which allowed DOS to use the extra memory in a 386 as EMS.
TBH, Windows/386 wasn't very popular or very widely-used. If you wanted the power of a 386 with DOS apps, then you probably were fine with or even preferred text-mode stuff and didn't want a GUI. Bear in mind this is long before graphics accelerators had been invented. Sure you could tile several DOS apps side-by-side, but then you could only see a little bit of each one -- VGA cards and monitors only supported 640×480 pixels. Windows 2 wasn't really established enough to have special hi-res superVGA cards available for it yet.*
Windows/386 could also multitask DOS apps full-screen, and if you used graphical DOS apps, you had to run them full-screen. Windows/386 couldn't run graphical DOS apps inside windows.
But if you used full-screen multitasking, with hotkeys instead of a mouse, then why not use something like DESQview anyway? It used way less disk and memory than Windows, and it was quicker and had no driver issues, because it didn't support any additional drivers.
The big mistake MS and IBM made when they wrote OS/2 was that they should have targeted the 386 chip, instead of the 286.
Microsoft knew this – it even had a prototype OS/2 1 for 386, codenamed "Sizzle" and "Football" – but IBM refused because when it sold thousands of 286 PS/2 machines it had promised the customers OS/2 for them. The customers didn't care, they didn't want OS/2, and this mistake cost IBM the entire PC industry.
If OS/2 1 had been a 386 OS it could have multitasked DOS apps, and PC power users would have been all over it. But it wasn't, it was a 286 OS, and it could only run 1 DOS app at a time. For that, the expensive upgrade and extra RAM you needed wasn't worth it.
So OS/2 bombed. Windows 2 bombed too. But MS was so disheartened by IBM's intransigence, it went back to the dead Windows 2 product, gave it a facelift with the look-and-feel
If you ran Windows 3 on an 8086 or a machine with only 640kB (i.e. no XMS), you got a Real Mode 8086-only GUI on top of DOS.
If you ran Win3 on a 286 with 1MB-1¾MB of RAM then it launched in Standard Mode and magically became a 16-bit DOS extender, giving you access to up to 16MB of RAM (if you were rich and
If you ran W3 on a 386 with 2MB of RAM or more, it launched in 386 Enhanced Mode and became a 32-bit multitasking DOS extender and could multitask DOS apps, give you virtual memory and a memory space of up to 4GB.
All in a single product on one set of disks.
This was revolutionary, and it was a huge hit...
And that about wrapped it up for OS/2.
Windows 3.0 was very unreliable and unstable. It often threw what it called an Unrecoverable Application Error (UAE) – which even led to a joke T-shirt that said "I thought UAE was a country in Arabia until I discovered Windows 3!"... but when it worked, what it did was amazing for 1990.
Microsoft eliminated UAEs in Windows 3.1, partly by a clever trick: it renamed the error to "General Protection Fault" (GPF) instead.
Me, personally, always the contrarian, I bought OS/2 2.0 with my own money and I loved it. It was much more stable than Windows 3, multitasked better, and could do way more... but Win3 had the key stuff people wanted.
Windows 3.1 bundled the separate Multimedia Extensions for Windows and made it a bit more stable. Then Windows for Workgroups bundled all that with networking, too!
Note — in the DOS era, all apps needed their own drivers. Every separate app needed its own printer drivers, graphics drivers (if it could display graphics in anything other than the standard CGA, EGA, VGA or Hercules modes), sound drivers, and so on.
One of WordPerfect's big selling points was that it had the biggest and best set of printer drivers in the business. If you had a fancy printer, WordPerfect could handle it and use all its special fonts and so on. Quite possibly other mainstream offerings couldn't, so if you ran WordStar or MultiMate or something, you only got monospaced Courier in bold, italic, underline and combinations thereof.
This included networking. Every network vendor had their own network stack with their own network card drivers.
And network stacks were big and each major vendor used their own protocol. MS used NetBEUI, Novell used IPX/SPX, Apple used AppleTalk, Digital Equipment Corporation's PATHWORKS used DECnet, etc. etc. Only weird, super-expensive Unix boxes that nobody could afford used TCP/IP.
You couldn't attach to a Microsoft server with a Novell network stack, or to an Apple server with a Microsoft stack. Every type of server needed its own unique special client.
This basically meant that a PC couldn't be on more than one type of network at once. The chance of getting two complete sets of drivers working together was next to nil, and if you did manage it, there'd be no RAM left to run any apps anyway.
Windows changed a lot of things, but shared drivers were a big one. You installed one printer driver and suddenly all your apps could print. One sound driver and all your apps could make noises, or play music (or if you had a fancy sound card, both!) and so on. For printing, Windows just sent your printer a bitmap — so any printer that could print graphics could suddenly print any font that came with Windows. If you had a crappy old 24-pin dot-matrix printer that only had one font, this was a big deal. It was slow and it was noisy but suddenly you could have fancy scalable fonts, outline and shadow effects!
But when Microsoft threw networking into this too, it was transformative. Windows for Workgroups broke up the monolithic network stacks. Windows drove the card, then Windows protocols spoke to the Windows driver for the card, then Windows clients spoke to the protocol.
So now, if your Netware server was configured for AppleTalk, say — OK, unlikely, but it could happen, because Macs only spoke AppleTalk — then Windows could happily access it over AppleTalk with no need for IPX.
The first big network I built with Windows for Workgroups, I built dual-stack: IPX/SPX and DECnet. The Netware server was invisible to the VAXen, and vice versa, but WfWg spoke to both at once. This was serious black magic stuff.
This is part of why, over the next few years, TCP/IP took off. Most DOS stuff never really used TCP/IP much — pre-WWW, very few of us were on the Internet. So, chaos reigned. WfWg ended that. It spoke to everything through one stack, and it was easy to configure: just point-and-click. Original WfWg 3.1 didn't even include TCP/IP as standard: it was an optional extra on the disk which you had to install separately. WfWg 3.11 included 16-bit TCP/IP but later Microsoft released a 32-bit TCP/IP stack, because by 1994 or so, people were rolling out PC LANs with pure IP.
* Disclaimer: this is a slight over-simplification for clarity, one of several in this post. A tiny handful of SVGA cards existed, most of which needed special drivers, and many of which only worked with a tiny handful of apps, such as one particular CAD program, or the GEM GUI, or something obscure. Some did work with Windows 2, but if they did, they were all-but unusable because Windows 2's core all had to run in the base 640kB of RAM and it very easily ran out of memory. Windows 3 was not much better, but Windows 3.1 finally fixed this a bit.
So if you had an SVGA card and Windows/286 or Windows/386 or even Windows 3.0, you could possibly set some super-hires mode like 1024×768 in 16 colours... and admire it for whole seconds, then launch a few apps and watch Windows crash and die. If you were in something insane like 24-bit colour, you might not even get as far as launching a second app before it died.
Clarification for the obsessive: when I said 1¾MB, that was also a simplification. The deal was this:
If you had a 286 & at least 1MB RAM, then all you got was Standard Mode, i.e. 286 mode. More RAM made things a little faster – not much, because Windows 2 didn't have a disk cache, relying on DOS to do that. If you had 2 MB or 4 or 8 or 16 (not that anyone sane would put 16MB in a 286, as it would cost $10,000 or something) it made no odds: Standard Mode was all a 286 could do.
If you had a 386 and 2MB or more RAM, you got 386 Enhanced Mode. This really flew if you had 4MB or more, but very few machines came with that much except some intended to be servers, running Unix of one brand or another. Ironically, the only budget 386 PC with 4MB was the Amstrad 2386, a machine now almost forgotten by history. Amstrad created the budget PC market in Europe with the PC1512 and PC1640, both 8086 machines with 5.25" disk drives.
It followed this with the futuristic 2000 series. The 2086 was an unusual PC – an ISA 8086 with VGA. The 2286 was a high-end 286 for 1988: 1MB RAM & a fast 12.5MHz CPU.
But the 2386 had 4MB as standard, which was an industry-best and amazing for 1988. When Windows 3.0 came out a couple of years later, this was the only PC already on the market that could do 386 Enhanced Mode justice, and easily multitask several DOS apps and big high-end Windows apps such as PageMaker and Omnis. Microsoft barely offered Windows apps yet – early, sketchy versions of Word and Excel, nothing else. I can't find a single page devoted to this remarkable machine – only its keyboard.
The Amstrad 2000 series bombed. They were premature: the market wasn't ready and few apps used DOS extenders yet. Only power users ran OS/2 or DOS multitaskers, and power users didn't buy Amstrads. Nor did people who wanted a server for multiuser OSes such as Digital Research's Concurrent DOS/386.
Its other bold design move was that Amstrad gambled on 5.25" floppies going away, replaced by 3.5" diskettes. They were right, of course – and so the 2000 series had no 5.25" bays, allowing for a sleek, almost aerodynamic-looking case. But Amstrad couldn't foresee that soon CD-ROM drives would be everywhere, then DVDs and CD burners, and the 5.25" bay would stick around for another few decades.