Obsolete processors, 286 vs. 386

[bitrex]

The article here states regarding the 386SX at 16MHz

"The 386 was a huge advance but you'd never know it from one of these
little slugs — they were usually out-performed by the better 286s."

"Even with 4MB or 8MB RAM, you wouldn't want to run Windows 3.1 on a
386SX-16 though. The SX-33s and DX-40s that followed soon after were
vastly faster."

<http://www.redhill.net.au/c/c-2.html>

So my family actually had one of the "slug"-based PCs in the early
1990s. I don't remember it being nearly as bad as the article makes it
out to be, particularly with respect to the important things in a 13
year old's life at that time: video games.

It claims some of the better 286es would perform as well in practice as
this processor. I had a friend who had the "standard issue" 286: 286-16
MHz, 1 meg RAM, 256k VGA card, 40 meg hard drive.

When playing the games of the time that relied heavily on "pseudo-3D"
CPU effects with a lot of sprite scaling, fixed-point math for
calculating angles the 386 system would run rings around the 286 system
- whatever graphics code it was that was unusably slow on the 286 was
nice and smooth on the 386. The article seems to be talking about
performance of business applications and maybe the comparison was valid
there, but for "leisure" applications there wasn't any comparison. It
was even better when the stock 1MB of RAM was upgraded to 2.

Wondering what might have made the difference; the article claims the
SX-16 didn't have an onboard cache but the 386 arch IIRC supported an
external cache; not sure how many systems actually implemented this.
Faster bus clock, maybe?

To my recollection Win 3.1 also ran fine on a 386SX-16 with 2 megs of
RAM and a 100 meg hdd. A few years later another friend's family picked
up a 486DX/2-66 which of course would smoke everything else we had
available.

[bitrex]

On 10/13/2017 07:29 AM, bitrex wrote:
> The article here states regarding the 386SX at 16MHz
>
> "The 386 was a huge advance but you'd never know it from one of these
> little slugs — they were usually out-performed by the better 286s."
>
> "Even with 4MB or 8MB RAM, you wouldn't want to run Windows 3.1 on a
> 386SX-16 though. The SX-33s and DX-40s that followed soon after were
> vastly faster."
>
> <http://www.redhill.net.au/c/c-2.html>
>
> So my family actually had one of the "slug"-based PCs in the early
> 1990s. I don't remember it being nearly as bad as the article makes it
> out to be, particularly with respect to the important things in a 13
> year old's life at that time: video games.
>
> It claims some of the better 286es would perform as well in practice as
> this processor. I had a friend who had the "standard issue" 286: 286-16
> MHz, 1 meg RAM, 256k VGA card, 40 meg hard drive.

It's possible I'm remembering incorrectly and it was actually a 12MHz 286.

[Jan Panteltje]

On a sunny day (Fri, 13 Oct 2017 07:29:44 -0400) it happened bitrex
<bit...@de.lete.earthlink.net> wrote in <Is1EB.2153$pU1....@fx17.iad>:

>The article here states regarding the 386SX at 16MHz
>
>"The 386 was a huge advance but you'd never know it from one of these

>little slugs â they were usually out-performed by the better 286s."

My first PC (I bought) was a 486DX2-66 running win3.1 on DRDOS
It replaced my home build Z80 system, but my Z80 was faster as it had a RAM disk that it loaded
the floppies in, so you would always work from RAM disk (was I/O mapped).

For my work I designed among other things ISA cards with stuff for that old 286 and 386.
I never liked the architecture, and still don't.
We worked close with IBM, I remember boss called me in his office one day
and showed the first 386, it WAS blazingly fast..
Games? I dunno, I had a 3D game with planes on that DX2-66, needed special shutter glasses, shutters driven from the parport.
When the evil Widows from RatMond came with win 98 (probably in 1998) that thing they dared sell as an OS
(was actually just to kill Digital Research DOS) no longer would run on DRDOS,
so and then my 3Dgame no longer worked either.
In the same year however I found Linux on some CD.. and that was the end for that Ratmond crap.

Now with raspberries and pads? who needs a PC?
;-)

[bitrex]

On 10/13/2017 07:59 AM, Jan Panteltje wrote:

> My first PC (I bought) was a 486DX2-66 running win3.1 on DRDOS
> It replaced my home build Z80 system, but my Z80 was faster as it had a RAM disk that it loaded
> the floppies in, so you would always work from RAM disk (was I/O mapped).
>
> For my work I designed among other things ISA cards with stuff for that old 286 and 386.
> I never liked the architecture, and still don't.
> We worked close with IBM, I remember boss called me in his office one day
> and showed the first 386, it WAS blazingly fast..
> Games? I dunno, I had a 3D game with planes on that DX2-66, needed special shutter glasses, shutters driven from the parport.
> When the evil Widows from RatMond came with win 98 (probably in 1998) that thing they dared sell as an OS
> (was actually just to kill Digital Research DOS) no longer would run on DRDOS,
> so and then my 3Dgame no longer worked either.
> In the same year however I found Linux on some CD.. and that was the end for that Ratmond crap.
>
> Now with raspberries and pads? who needs a PC?
> ;-)
>

Unfortunately there's still a lot of stuff that's not built for ARM.

[upsid...@downunder.com]

The 80286 was just a stop gap between 8086 and iAPX432 when Intel
realized that they could not get decent performance out of the 432.

The segment system of 286 was stupid compared to later x86 models. In
addition, you could switch the 286 from normal mode to protected mode,
but switching back required setting up the reset vector to a DOS
program and performing a processor reset to run a DOS program.

I acquired a 10 MHz 286 when I started my company, mainly for sending
invoices and using it as a terminal emulator with disk storage.

There was a semi-graphical Windows 2.x running on 286, but more or
less useless for any real work. It was much more convenient to use
plain MS-DOS.

In retrospect I was lucky, when I bought the next computer (486) and
invested in a sufficient large memory for running Windows NT 3.51
instead of wafting for Win95 introduction later that year. After
reading horror stories about Win9x unreliability, NT3.51 was as stable
as RSX-11 or VMS that I had been used to in the previous decades.

[Terje Mathisen]

The article is simply bogus:
The first available 386 PC was a 16 MHz model from Compaq and cost a
small fortune, but it was a huge step forward compared to the
(overclocked) 286 machines that preceeded it.

Terje

--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"

[Mark Storkamp]

In article <Is1EB.2153$pU1....@fx17.iad>,

My first (Microsoft compatible) PC was a 33Mhz 386 with a 64K direct
mapped cache. I was one of the first on the block to get one since I
just started working for the company making the chip sets. It ran rings
around anything else that was available at the time, but there were
initial problems with getting the cache working, and until it did it was
less than impressive. Prior to that job I worked with a Data General
Eclipse MV/8000, and going to the new job felt like I had stepped 10
years back in time. It took a while to get used to the idea I couldn't
just start a new thread running, or have a background task working on
something. I think we could have been much further along much faster if
Microsoft never existed.

[Jan Panteltje]

On a sunny day (Fri, 13 Oct 2017 09:12:11 -0400) it happened bitrex
<bit...@de.lete.earthlink.net> wrote in <LY2EB.2154$pU1....@fx17.iad>:

It was a bit of making fun, while writing that I was using 3 raspies, 1 PC, 2 laptops.
;-)

But of course you need to compile from source, so all that closed source stuff is out.

[Quadibloc]

On Friday, October 13, 2017 at 5:29:47 AM UTC-6, bitrex wrote:

> To my recollection Win 3.1 also ran fine on a 386SX-16 with 2 megs of
> RAM and a 100 meg hdd.

That agrees with my recollection as well.

John Savard

[Anne & Lynn Wheeler]

I was in skirmish with Boca regarding the nature of the PC market. I
would periodically post on internal formers, single unit prices from
Sunday SJMN ... way below Boca predictions. Note late 80s, makers on the
other side of the pacific built up huge inventory of 286 machines for
the holiday market ... then intel announces 386sx (significantly reduces
system cost with lots of stuff incorporated into 386 processor chip) ...
and there are enormous 286 fire sales.

Then head of Boca does large contract with Dataquest (since bought by
Gartner) to do detailed study and future of PC market ... that calls for
several hour video tape of silicon valley experts. I know the person
running the Dataquest study and am asked to be one of the silicon valley
experts ... I clear it with my management and Dataquest promises to
garble my bio/intro so Boca won't recognize me.

recent post with my last post on internal forum (july 1992), just days
before leaving IBM
http://www.garlic.com/~lynn/2017i.html#0

--
virtualization experience starting Jan1968, online at home since Mar1970

[bitrex]

An accurate monday-morning-quarterback prediction in 1992 for circa 1997
regarding still using a top of the line 386: your friends would laugh at
you and you wouldn't be able to give it away.

[Melzzzzz]

On 2017-10-13, bitrex <bit...@de.lete.earthlink.net> wrote:
> The article here states regarding the 386SX at 16MHz
>
> "The 386 was a huge advance but you'd never know it from one of these
> little slugs — they were usually out-performed by the better 286s."

False. 386 SX was 32 bit, same as 386DX, but with 16 bit bus.

--
press any key to continue or any other to quit...

[Michael A. Terrell]

upsid...@downunder.com wrote:
>
> There was a semi-graphical Windows 2.x running on 286, but more or
> less useless for any real work. It was much more convenient to use
> plain MS-DOS.

We had a PC running Windows 2.0 as the controller for a SATE test
system at Microdyne. It was designed and built by Scientific Atlanta
when they copied the design of one of our telemetry products, and put it
into another case. The only real difference was that they used a LCD
display instead of LEDs, but they copied several Microdyne patented
components.

We came into work one Monday morning to discover that the IT
characters had wiped the hard drive, and installed Win 95 in ts place.
They didn't back up the custom software, or the OS. They caused us to
lose a week's production on that product line. After that, they had to
ask permission to touch any non networked computer in the company.


--
Never piss off an Engineer!

They don't get mad.

They don't get even.

They go for over unity! ;-)

[timca...@aol.com]

For many years my rule of thumb was that 1 computer year == 10 human years.
What this meant:
First year a CPU was introduced, it was very expensive and probably not worth it.
Second year (the teenage years :) ) it might be worth it.
Third year was usually the peak of price/performance.
Between the 6th and 7th year, the CPU was due for retirement, but could still be used.
By the 10th year, it was junk.

So, 386-16 introduced in 1986, barely usable by 1992, dead by 1996.
486DX25 introduced 1989, peaked in 1992, dead by 1999.
Pentium 60, release 1993, usable up 1999/2000, dead by 2003.

Anyway, this worked up until about 2010, then things slowed way down.

- Tim

[Jasen Betts]

On 2017-10-13, bitrex <bit...@de.lete.earthlink.net> wrote:

> The article here states regarding the 386SX at 16MHz
>
> "The 386 was a huge advance but you'd never know it from one of these
> little slugs — they were usually out-performed by the better 286s."
>
> "Even with 4MB or 8MB RAM, you wouldn't want to run Windows 3.1 on a
> 386SX-16 though. The SX-33s and DX-40s that followed soon after were
> vastly faster."
>
><http://www.redhill.net.au/c/c-2.html>
>
> So my family actually had one of the "slug"-based PCs in the early
> 1990s. I don't remember it being nearly as bad as the article makes it
> out to be, particularly with respect to the important things in a 13
> year old's life at that time: video games.
>
> It claims some of the better 286es would perform as well in practice as
> this processor. I had a friend who had the "standard issue" 286: 286-16
> MHz, 1 meg RAM, 256k VGA card, 40 meg hard drive.

a 286 with a hot-rod coprocessor "cyrix?" could easily beat a bare 386
(witn no floating point co-processor) at floating point tasks. most games
didn't use floating point back then, stuff like spreadsheets and spice
did.

--
This email has not been checked by half-arsed antivirus software

[Jasen Betts]

Not ALL of it, you can run the pi version of mathematica for free. The
H.264 codec too.

[jacko]

Too much was made of binary compatibility and not so clever self modifying code, instead of just replacing each byte by a 32 bit word, and having it as a jump table.

[MitchAlsup]

On Friday, October 13, 2017 at 4:25:54 PM UTC-5, timca...@aol.com wrote:
>
> So, 386-16 introduced in 1986, barely usable by 1992, dead by 1996.
> 486DX25 introduced 1989, peaked in 1992, dead by 1999.
> Pentium 60, release 1993, usable up 1999/2000, dead by 2003.

I had a 33 Mhz 486 machine, running windows 3.11, got tired of the the
constant swapping, and changed from 4 MB to 20MB. This was too much for
W 3.11 so I used the upper 4 MB as a disk cache and 16 MB as main memory.

A few years later, my wife got a 66 MHz 586 with 8MB, and my lowly 486
would wipe the floor in comparison to her machine. Two things made the
lower performing machine run better, 486 had more main memory, and second
it had 2 disks, one serving as the swap partition. My workload was more
difficult than hers .....

YMMV.

[George Neuner]

On Fri, 13 Oct 2017 15:02:41 -0700 (PDT), MitchAlsup
<Mitch...@aol.com> wrote:

>I had a 33 Mhz 486 machine, running windows 3.11, got tired of the the
>constant swapping, and changed from 4 MB to 20MB. This was too much for
>W 3.11 so I used the upper 4 MB as a disk cache and 16 MB as main memory.
>
>A few years later, my wife got a 66 MHz 586 with 8MB, and my lowly 486
>would wipe the floor in comparison to her machine. Two things made the
>lower performing machine run better, 486 had more main memory, and second
>it had 2 disks, one serving as the swap partition. My workload was more
>difficult than hers .....
>
>YMMV.

FWIW:

It was fairly well known back then that you could get a significant
performance boost for ALU code on the P5 by compiling for the i486
instead.

It had to do with the ratio of "simple" vs "complex" instructions in
the code. The P5's ALU pipelines were assymetric: 'A' could handle
any ALU instruction, but 'B' could handle only REG->REG instructions.
The more simple instructions, the more the pipelines could be used in
parallel.

A number of mainstream C/C++ compilers [notably Microsoft's, but there
were others] preferred to use complex instructions when told
specifically to compile for P5 ... basically treating it like a fast
i386. But when told to compile for the i486, they preferred sequences
of simpler instructions. i486 specific code typically was a bit
larger than P5 specific code, but often ran ~10% faster.

George

[Jeremy Linton]

On 10/13/2017 6:29 AM, bitrex wrote:
> The article here states regarding the 386SX at 16MHz
>
> "The 386 was a huge advance but you'd never know it from one of these
> little slugs — they were usually out-performed by the better 286s."
>
> "Even with 4MB or 8MB RAM, you wouldn't want to run Windows 3.1 on a
> 386SX-16 though. The SX-33s and DX-40s that followed soon after were
> vastly faster."
>
> <http://www.redhill.net.au/c/c-2.html>
>
> So my family actually had one of the "slug"-based PCs in the early
> 1990s. I don't remember it being nearly as bad as the article makes it
> out to be, particularly with respect to the important things in a 13
> year old's life at that time: video games.
>
> It claims some of the better 286es would perform as well in practice as
> this processor. I had a friend who had the "standard issue" 286: 286-16
> MHz, 1 meg RAM, 256k VGA card, 40 meg hard drive.
>
> When playing the games of the time that relied heavily on "pseudo-3D"
> CPU effects with a lot of sprite scaling, fixed-point math for
> calculating angles the 386 system would run rings around the 286 system
> - whatever graphics code it was that was unusably slow on the 286 was
> nice and smooth on the 386. The article seems to be talking about
> performance of business applications and maybe the comparison was valid
> there, but for "leisure" applications there wasn't any comparison. It
> was even better when the stock 1MB of RAM was upgraded to 2.

I remember having these discussions in school, along with instruction
timing tables. There were quite a number of "critical" instructions
which were much slower on the 386. In/Out is the one I remember.

http://zsmith.co/intel_i.html#in

The biggest problem IIRC was the price difference, you could get a lot
higher clocked 286 than 386 for less money for quite a while, and the
extra 4+ Mhz made all the difference.

So, it ended up being a wash, code optimized for a 386 was faster, but a
lot of stuff, particularly older games the 286 was faster. Which of
course doesn't jive with your memory. But the problem IIRC with the SX
was it added an addition penalty for a lot of the native 386 code due to
the 16-bit bus. Meaning that it did miserable vs a "Real" 386 on 386
only code. I seem to remember running win3.0 286 mode a lot because it
was faster than full 386 mode but still allowed me to access more ram
than the real mode. When win 3.1 came out that was one of the things
they removed forcing me to run either real mode or 386 enhanced mode...

Also, I was late to the 32-bit party because 286's were so much less
expensive than 386's (even following the 386sx) that it was much cheaper
to buy a higher clocked 286 (20's and 25's) which out performed the
basic 386's 12's and 16's.

It was pretty much the same problem when I finally got a 486, I remember
having a hard time deciding between a 386-40, and a low end 486. When I
finally upgraded it was to a 486SX-20, which came "overclocked" from my
reseller @ 40Mhz. That machine with a MR-BIOS motherboard and VLB video
card was such a huge upgrade for so little money that I turned around
and sold maybe a dozen of them to friends of mine.

[George Neuner]

On Fri, 13 Oct 2017 19:40:46 +0000 (UTC), Melzzzzz
<Melz...@zzzzz.com> wrote:

>On 2017-10-13, bitrex <bit...@de.lete.earthlink.net> wrote:
>> The article here states regarding the 386SX at 16MHz
>>
>> "The 386 was a huge advance but you'd never know it from one of these
>> little slugs — they were usually out-performed by the better 286s."
>
>False. 386 SX was 32 bit, same as 386DX, but with 16 bit bus.

It isn't false.

The only time an i386SX really could outperform an equivalent clocked
i286 was running 16-bit code that did a lot of multiword (32-bit or
wider) computation, or code that did a whole lot of segment switching
[which was significantly faster on the i386 than on the i286].

But most people would not have discovered this - apart from some
enormous WYSIWYG page design apps, there just weren't many 16-bit
programs that would run faster on a i386SX unless the clock rate also
was significantly faster.

And the i386SX was abysmal when running 32-bit code ... on average
only ~40% the performance of a DX at the same clock speed.

George

[George Neuner]

On 13 Oct 2017 21:08:19 GMT, Jasen Betts <ja...@xnet.co.nz> wrote:

>a 286 with a hot-rod coprocessor "cyrix?" could easily beat a bare 386
>(witn no floating point co-processor) at floating point tasks. most games
>didn't use floating point back then, stuff like spreadsheets and spice
>did.

No "hot rods" necessary ... an i286 with i287 coprocessor could
roughly equal an i386DX running x87 emulation at 1/3 higher clock
rate.

Cyrix's 82S87 was compatible with the i287, but ~30-40% faster.
Weitek's 1067 was not i287 compatible, and IIRC handled only 32-bit
floats and no trancendentals, but was 5-6x faster than the i287.

George

[bitrex]

The difference that I recall then may be because I'm not remembering
entirely correctly - my buddy's 286 was more likely an 8 or 12 MHz model
and my machine's performance advantage was simply from raw clock advantage.

[Jeremy Linton]

This is basically what I remember, plus the fact that if you had a fixed
amount of money, you could get far better performance with 286-20/24
than a 386SX at 16Mhz.


>
> George
>

[Jeremy Linton]

This is changing very rapidly for linux/opensource projects. Most of the
core infrastructure is in place (kvm, xen, docker, etc). So much so that
all the mainstream distro's are basically building the same packages for
x86-64 as arm64 at this point. That includes qemu on arm64, which can
emulate just about anything you might want.

Or you can just run your x86 binaries on windows/arm64 directly...

http://www.zdnet.com/article/microsofts-x86-on-arm64-emulation-a-windows-10-redstone-3-fall-2017-deliverable/

[k...@notreal.com]

No, the '286 was the stopgap between the 8086 and the '386.
The '386 was architected first but it couldn't be built. The 432 was
a whole different kettle of rotting fish.

>The segment system of 286 was stupid compared to later x86 models. In
>addition, you could switch the 286 from normal mode to protected mode,
>but switching back required setting up the reset vector to a DOS
>program and performing a processor reset to run a DOS program.

You weren't supposed to switch back. That defeated the memory
protection (which no one used).

[Joe Pfeiffer]

I've found over many decades, that the second processor from the top in
specifications is almost certain to be the best for price-performance.

[Melzzzzz]

I didn't know that at all. I was pretty happy switching from 286 to
386SX ;)
I can't remember what clock was running but I think it was more then
16Mhz. What I needed most 386 for is to install SCO Unix as that
didn't work on 286.

[Robert Wessel]

While it's certainly true that a 386SX was a bit slower than an
equivilently clocked 286, most of the time it wasn't a big difference.
OTOH, while many 386SX shipped at 16MHz, there were many 20 and even
24MHz 286 clones out there (I don't know if Intel even shipped a 286
faster than 12MHz, but they sure weren't common)..

OTOH, the 386SX had a couple of massive advantages over the 286: the
ability to have (E)EMS memory without special hardware, and the
ability to run Windows, or other DOS multi-tasker, in 386 mode,
instead of using the 286 hack. Even if you were using DOS,
Expanded/Extended memory for disk cache and for holding a chunk of
your network stack was a big win.

[George Neuner]

On Fri, 13 Oct 2017 23:04:20 -0500, Robert Wessel
<robert...@yahoo.com> wrote:

>While it's certainly true that a 386SX was a bit slower than an
>equivilently clocked 286, most of the time it wasn't a big difference.
>OTOH, while many 386SX shipped at 16MHz, there were many 20 and even
>24MHz 286 clones out there (I don't know if Intel even shipped a 286
>faster than 12MHz, but they sure weren't common)..

I think you're right about that ... the i286 was available in 16, 20
and 25 MHz, but AFAICR those chips all were clones.

>OTOH, the 386SX had a couple of massive advantages over the 286: the
>ability to have (E)EMS memory without special hardware, and the
>ability to run Windows, or other DOS multi-tasker, in 386 mode,
>instead of using the 286 hack. Even if you were using DOS,
>Expanded/Extended memory for disk cache and for holding a chunk of
>your network stack was a big win.

No question there were valid reasons for wanting an i386SX over an
i286 ... it's just that performance/price wasn't one of them. <grin>

Anyone with 32-bit aspirations went for a DX unless money was very
tight.

George

[Jan Panteltje]

On a sunny day (13 Oct 2017 21:12:37 GMT) it happened Jasen Betts
<ja...@xnet.co.nz> wrote in <orra85$1tq$2...@gonzo.alcatraz>:

Yes, true I have Mathematica on one Pi, and even used it.

But there are open source alternatives for it IIRC (cannot remember the name).

But ffmpeg compiles with H264 does it not?

From ffmpeg -codecs:
decode H264
D V D h264 H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10
encode to 264???
EV libx264 libx264 H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10
EV libx264rgb libx264 H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 RGB

Works OK here on PC :-)
/root/compile/ffmpeg/ffmpeg-0.11.1/libavcodec/libx264.c

??
Would that not compile or Raspberry?

(got to it try some day).
And mplayer links against ffmpeg, or has it in its source.

[Martin Brown]

On 14/10/2017 02:44, George Neuner wrote:
> On Fri, 13 Oct 2017 19:40:46 +0000 (UTC), Melzzzzz
> <Melz...@zzzzz.com> wrote:
>
>> On 2017-10-13, bitrex <bit...@de.lete.earthlink.net> wrote:
>>> The article here states regarding the 386SX at 16MHz
>>>
>>> "The 386 was a huge advance but you'd never know it from one of these
>>> little slugs — they were usually out-performed by the better 286s."
>>
>> False. 386 SX was 32 bit, same as 386DX, but with 16 bit bus.
>
> It isn't false.

It depends what you were trying to do and under what operating system.
Running DOS and 16bit code I suspect the fastest 286's could beat a
stock 386SX but running OS/2 or 32 bit code the situation was reversed.

IBM's decision to make OS/2 "run" (more like limp) on the 286 was one of
the worst decisions they ever made. That and conflating it with PS/2 the
attempt to lock in buyers to a new proprietary MCA bus architecture.

That mistake quite literally opened the window for Microsoft Windows.

> The only time an i386SX really could outperform an equivalent clocked
> i286 was running 16-bit code that did a lot of multiword (32-bit or
> wider) computation, or code that did a whole lot of segment switching
> [which was significantly faster on the i386 than on the i286].

OS/2 or the various dialects of Unix available at the time.

> But most people would not have discovered this - apart from some
> enormous WYSIWYG page design apps, there just weren't many 16-bit
> programs that would run faster on a i386SX unless the clock rate also
> was significantly faster.
>
> And the i386SX was abysmal when running 32-bit code ... on average
> only ~40% the performance of a DX at the same clock speed.

But it would run it and a fair bit faster than the 286 for the same
compiled program code too. We always used machines with the numeric
coprocessor fitted which made a huge difference for scientific computing
(and later the Cyrix FASMATH when that became available).

It always amused me back then that the COCOM rules made it a hanging
offence to export Compaq PCs to the USSR whilst IBM PCs were permitted.

--
Regards,
Martin Brown

[bitrex]

Looking at the 286-386-486 instruction set it looks like the biggest
difference is that multiply instructions have the opportunity to take
significantly fewer clock cycles on the 386 than on the 286, where
multiplies seem to take a fixed number of cycles (~20.)

It was games that involved a lot of rotations/scaling where I noticed a
big difference. 2D graphic performance seemed about the same.

[Phil Hobbs]

That was before L2 cache appeared in micros, so it took twice as many
main memory accesses just for instruction fetch.

The 286 _could_ be switched from protected mode back to real mode, but
only by hitting RST or NMI or something like that--I forget. MS Xenix
and IBM/MS OS/2 1.x ran on '286es.

286es were also really slow at ring transitions, so as a performance
optimization the OS/2 architects permitted direct port I/O from user
code at Ring 2. For reasons of backward compatibility, that still
worked in 3.x and 4.x, but only from a 16-bit code segment with I/O
privilege level (IOPL) set to 2. I did a number of instruments with
bidirectional parallel-port I/O that way. (I built a 16-bit IOPL DLL
once and used it for a long time.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

[Tim Williams]

"Phil Hobbs" <pcdhSpamM...@electrooptical.net> wrote in message
news:Uu6dndcQ3syn53_E...@supernews.com...

> The 286 _could_ be switched from protected mode back to real mode, but
> only by hitting RST or NMI or something like that--I forget. MS Xenix and
> IBM/MS OS/2 1.x ran on '286es.

IIRC, the initial hack was inducing a triple fault, but first, politely
setting a flag in the BIOS RAM area that it's not a cold or warm boot, but a
cup of coffee and a "tell me where to get back to things" state switch. :^)

Then they later added RST or NMI via KB controller, unless I'm confounding
that with the A20 hack, which was yet another stupid 286+ real mode hack.

Tim

--
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com

[Robert Wessel]

On Sat, 14 Oct 2017 23:49:53 -0500, "Tim Williams"
<tmor...@gmail.com> wrote:

>"Phil Hobbs" <pcdhSpamM...@electrooptical.net> wrote in message
>news:Uu6dndcQ3syn53_E...@supernews.com...
>> The 286 _could_ be switched from protected mode back to real mode, but
>> only by hitting RST or NMI or something like that--I forget. MS Xenix and
>> IBM/MS OS/2 1.x ran on '286es.
>
>IIRC, the initial hack was inducing a triple fault, but first, politely
>setting a flag in the BIOS RAM area that it's not a cold or warm boot, but a
>cup of coffee and a "tell me where to get back to things" state switch. :^)
>
>Then they later added RST or NMI via KB controller, unless I'm confounding
>that with the A20 hack, which was yet another stupid 286+ real mode hack.

I think that's backwards. The original implementation was the reset
via the (very slow) keyboard controller. The later discovered "hack"
was that a triple fault would cause a reset much more quickly, since
the resulting shutdown cycle on the CPU bus caused the motherboard
hardware to issue a reset.

In either case, the reset handler code would look at some specific
items in storage to determine if this was a "real" reset, or one that
was intended to just switch back to real mode.

[Kalle Olavi Niemitalo]

Jasen Betts <ja...@xnet.co.nz> writes:

> a 286 with a hot-rod coprocessor "cyrix?" could easily beat a bare 386
> (witn no floating point co-processor) at floating point tasks. most games
> didn't use floating point back then, stuff like spreadsheets and spice
> did.

Scorched Earth sure got faster when my father installed a Cyrix
FasMath coprocessor in our 286 box.

There also was an odd side effect: on each boot thereafter, the
border of the screen would briefly turn from black to some color.
I still don't understand why. The color was not the same one
each time, so it did not seem a deliberate feature of the BIOS.

[Tauno Voipio]

On 15.10.17 07:49, Tim Williams wrote:
> "Phil Hobbs" <pcdhSpamM...@electrooptical.net> wrote in message
> news:Uu6dndcQ3syn53_E...@supernews.com...
>> The 286 _could_ be switched from protected mode back to real mode, but
>> only by hitting RST or NMI or something like that--I forget.  MS Xenix
>> and IBM/MS OS/2 1.x ran on '286es.
>
> IIRC, the initial hack was inducing a triple fault, but first, politely
> setting a flag in the BIOS RAM area that it's not a cold or warm boot,
> but a cup of coffee and a "tell me where to get back to things" state
> switch. :^)
>
> Then they later added RST or NMI via KB controller, unless I'm
> confounding that with the A20 hack, which was yet another stupid 286+
> real mode hack.
>
> Tim

The A20 hack was there to support 8086 / 8088 code using a common
segment register for ROM at top of the 1 MiB address space and
RAM at the very bottom. It allowed an address overflow to wrap
around to the bottom.

The address hack was against the usage recommendations by Intel
from the very first 8086 family manuals. It was not the only
design decision by IBM which was completely wrong. Another was
to use interrupt numbers in the range 0 - 31 for the BIOS, though
Intel warned against such use. This made it impossible to run
base DOS in any of the 80186 family processors.

--

-TV

[Megol]

On Sunday, October 15, 2017 at 7:14:46 AM UTC+2, robert...@yahoo.com wrote:
> On Sat, 14 Oct 2017 23:49:53 -0500, "Tim Williams"
> <tmor...@gmail.com> wrote:
>
> >"Phil Hobbs" <pcdhSpamM...@electrooptical.net> wrote in message
> >news:Uu6dndcQ3syn53_E...@supernews.com...
> >> The 286 _could_ be switched from protected mode back to real mode, but
> >> only by hitting RST or NMI or something like that--I forget. MS Xenix and
> >> IBM/MS OS/2 1.x ran on '286es.
> >
> >IIRC, the initial hack was inducing a triple fault, but first, politely
> >setting a flag in the BIOS RAM area that it's not a cold or warm boot, but a
> >cup of coffee and a "tell me where to get back to things" state switch. :^)
> >
> >Then they later added RST or NMI via KB controller, unless I'm confounding
> >that with the A20 hack, which was yet another stupid 286+ real mode hack.
>
>
> I think that's backwards. The original implementation was the reset
> via the (very slow) keyboard controller. The later discovered "hack"
> was that a triple fault would cause a reset much more quickly, since
> the resulting shutdown cycle on the CPU bus caused the motherboard
> hardware to issue a reset.

Don't forget having a dedicated IO port for doing resets. Don't know if
that was a standardized port or chipset specific but it was there at
least for some systems. Took a long time for me to understand why one
would want something like that.

> In either case, the reset handler code would look at some specific
> items in storage to determine if this was a "real" reset, or one that
> was intended to just switch back to real mode.

The real fun was abusing that mechanism to create resident code. In
some cases one could even power-cycle the computer and the code would
still be running.

[Robert Wessel]

Actually that wasn't the issue with the 186. All later processors had
the same issues with IBM BIOS interrupts being used by the CPU. In
real mode the answer was the same in all cases, never do anything that
might trigger one of the interrupts that were (ab)used as the BIOS
API.

The 186 had an issue in that there were a bunch of integrated
peripherals, but they were not PC compatible. There were a number of
(mostly) PC compatible systems that did use the 186, but they had to
disable all the internal I/O. There were also some non-PC compatible
MS-DOS systems with the 186.

[Robert Wessel]

On Sun, 15 Oct 2017 07:04:08 -0700 (PDT), Megol <gole...@gmail.com>
wrote:

>On Sunday, October 15, 2017 at 7:14:46 AM UTC+2, robert...@yahoo.com wrote:
>> On Sat, 14 Oct 2017 23:49:53 -0500, "Tim Williams"
>> <tmor...@gmail.com> wrote:
>>
>> >"Phil Hobbs" <pcdhSpamM...@electrooptical.net> wrote in message
>> >news:Uu6dndcQ3syn53_E...@supernews.com...
>> >> The 286 _could_ be switched from protected mode back to real mode, but
>> >> only by hitting RST or NMI or something like that--I forget. MS Xenix and
>> >> IBM/MS OS/2 1.x ran on '286es.
>> >
>> >IIRC, the initial hack was inducing a triple fault, but first, politely
>> >setting a flag in the BIOS RAM area that it's not a cold or warm boot, but a
>> >cup of coffee and a "tell me where to get back to things" state switch. :^)
>> >
>> >Then they later added RST or NMI via KB controller, unless I'm confounding
>> >that with the A20 hack, which was yet another stupid 286+ real mode hack.
>>
>>
>> I think that's backwards. The original implementation was the reset
>> via the (very slow) keyboard controller. The later discovered "hack"
>> was that a triple fault would cause a reset much more quickly, since
>> the resulting shutdown cycle on the CPU bus caused the motherboard
>> hardware to issue a reset.
>
>Don't forget having a dedicated IO port for doing resets. Don't know if
>that was a standardized port or chipset specific but it was there at
>least for some systems. Took a long time for me to understand why one
>would want something like that.

I've heard of that, but I don't think I ever encountered it, at least
to my knowledge. The reset via the keyboard controller was via a
sequence of I/O port commands. I'm not sure how much advantage it
would have had over the triple-fault scheme.

[wolfgang kern]

"bitrex" polled for our experience ?
my first experience with PC/XT (better say first encounter) was about 1978.
I started and produced Z80 based mini-computers 1979.

my first bought PC was an Intel 286 (end 1982).
I tuned it up to to 20 MHz and 2.5 MB RAM (dont ask for the price now).
And I promoted and used AMD since 1979 for their reliable EPROMs.

it worked (somehow) with Windoze 3.11 (the first which worked at all)
and also WORD, Lotus123, Havard Graphics were often seen to be used.

Then the 386 store began..
My meanwhile well known salesmen offered me an hardware upgrade from
i286 to i386 for a "tiny" fee of 1000 AS (~90 US$ than).
And I offered him in return a gift of 1000 AS if he keep that shit !:)

So in short, I skipped the whole 386 aera and bougth myself the first
available 86_486 (an AMD of course) for my next PC.

Viewing back my to my decision seem to be nothing wrong at all.

after early AMD_486, I had K7, K8, PhenomII and Bulldoze now.
Not sure if my next generation, personal (for my tools) or commercial
(The things I sell) will be ZEN/RYZEN or just one step below this.

regardless of temporary lower ratings for AMD over Intel,....
I still prefer, promote and use AMD.
And not just because the performance I get per U$.
For me It's a only a deal of trustworthy vs. MacGreedy Ass-Creepers.
__
wolfgang (the elder)

[Gerhard Hoffmann]

Am 15.10.2017 um 17:27 schrieb Robert Wessel:
>

>
> Actually that wasn't the issue with the 186. All later processors had
> the same issues with IBM BIOS interrupts being used by the CPU. In
> real mode the answer was the same in all cases, never do anything that
> might trigger one of the interrupts that were (ab)used as the BIOS
> API.
>
> The 186 had an issue in that there were a bunch of integrated
> peripherals, but they were not PC compatible. There were a number of
> (mostly) PC compatible systems that did use the 186, but they had to
> disable all the internal I/O. There were also some non-PC compatible
> MS-DOS systems with the 186.
>

completely correct.

regards, Gerhard

[George Neuner]

On Sat, 14 Oct 2017 10:03:46 +0100, Martin Brown
<'''newspam'''@nezumi.demon.co.uk> wrote:

>On 14/10/2017 02:44, George Neuner wrote:
>
>> The only time an i386SX really could outperform an equivalent clocked
>> i286 was running 16-bit code that did a lot of multiword (32-bit or
>> wider) computation, or code that did a whole lot of segment switching
>> [which was significantly faster on the i386 than on the i286].
>
>OS/2 or the various dialects of Unix available at the time.

I never saw OS/2 on the i286, but I did have a chance to play a bit
with 286 Xenix and it was not bad at all. The system I saw basically
was running only office software - word processing, scheduling, etc. -
on smart terminals ... but it was supporting 6 simultaneous users.

>> But most people would not have discovered this - apart from some
>> enormous WYSIWYG page design apps, there just weren't many 16-bit
>> programs that would run faster on a i386SX unless the clock rate also
>> was significantly faster.
>>
>> And the i386SX was abysmal when running 32-bit code ... on average
>> only ~40% the performance of a DX at the same clock speed.
>
>But it would run it and a fair bit faster than the 286 for the same
>compiled program code too.

Well ... yeah ... because the i286 couldn't execute 32-bit code.
Pathethic is faster than not at all. <grin>

George

[timca...@aol.com]

I think you're dates are messed up:

IBM PC - Released Aug. 1981, but you couldn't really get one for awhile (6 to 9 months).
PC/XT - Released March 1983.
PC/AT (286) - Came out 1984 (but you couldn't get one for awhile)
Compaq 386 - Aug. 1986 (I think).

[Quadibloc]

On Monday, October 16, 2017 at 5:32:58 PM UTC-6, timca...@aol.com wrote:

> IBM PC - Released Aug. 1981, but you couldn't really get one for awhile (6 to 9 months).
> PC/XT - Released March 1983.
> PC/AT (286) - Came out 1984 (but you couldn't get one for awhile)
> Compaq 386 - Aug. 1986 (I think).

However, there were clone PCs that used an 8086 instead of an 8088 fairly early
on. And there were 5 megabyte accessory hard disk drives available for the PC
before IBM came out with the XT with a 10 megabyte drive included.

Intel released the 80286 on February 1st, 1982. So it's not _strictly_
impossible that someone could have laid his hands on an 80286-based PC clone in
1982. It is more likely that he did not remember his dates correctly, true.

John Savard

[Quadibloc]

On Monday, October 16, 2017 at 5:32:58 PM UTC-6, timca...@aol.com wrote:

> I think you're dates are messed up:

> IBM PC - Released Aug. 1981, but you couldn't really get one for awhile (6 to 9
> months).

Definitely he could not have first encountered the XT in 1978, so indeed his dates
are mixed up *there*. Of course, CP/M resembled PC-DOS so closely that confusion
is perhaps understandable.

John Savard

[wolfgang kern]

Could be that my memory from 35 years ago faded and I don't have the receipt
anymore. But right, my first AT was an IBM clone with an Intel 286 CPU and it
had a space-shuttle in the logo (pioneer?).
It came with DOS2.0 on several 180/360K 5.25" FDs, it had a 'large' 2MB HD and
the CPU was on an ISA card which later could be replaced by a 386 card.
__
wolfgang

[paul wallich]

On 10/17/17 12:00 AM, Quadibloc wrote:
> On Monday, October 16, 2017 at 5:32:58 PM UTC-6, timca...@aol.com wrote:
>
>> IBM PC - Released Aug. 1981, but you couldn't really get one for awhile (6 to 9 months).
>> PC/XT - Released March 1983.
>> PC/AT (286) - Came out 1984 (but you couldn't get one for awhile)
>> Compaq 386 - Aug. 1986 (I think).
>
> However, there were clone PCs that used an 8086 instead of an 8088 fairly early
> on. And there were 5 megabyte accessory hard disk drives available for the PC
> before IBM came out with the XT with a 10 megabyte drive included.

FWIW, there were 8086-based machines in (low-volume) production years
before the PC -- the 8088 was just there for cost reduction. So spinning
an 8086 motherboard would have been easy and a simple way to higher
performance (although compatibility might have been fun.

paul

[Chris]

Except that cpm was a much better sorted os, with sources included and
extensive docs on how to port of other machines and io devices. 16 bit
version and one for 68K and still don't understand why IBM chose
Microsoft, or Intel, for that matter

First pc here was a Tandon PCA/20, iirc, 286, 1 meg ram, 20 meg hard
drive, but had a 5 meg hard drive on an Apple II around 1984. Luxury.
Programming assembler etc, never filled it...

Regards,

Chris

[Bruce Hoult]

Well, you know ... Kildall was out flying, or wouldn't sign an NDA for a meeting he didn't know the purpose of (or possibly even the identity of the participants). And the Microsoft guys wore suits, and promised they had something they didn't have.

[Robert Wessel]

A number of folks built IBM PC compatibles with 8086, not least IBM
itself (some models of the PS/2 25 and 30 were 8086).

[Terje Mathisen]

Are you sure?

I would almost be willing to bet that the 80186 would have been the
alternative to the 286 if they wanted a lower-end option at that point
in time.

Terje

--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"

[Megol]

Yes there were - but not being IBM PC compatible and not running MSDOS.
They were also pretty expensive IIRC (but it was a long time ago).
--
My first x86 machine was the Wang PC which wasn't PC compatible, at least without an optional emulation board.
But it was advanced for its time, the standard text mode was 80x25
characters but using 10x12 pixel - much nicer than the then standard
8x8 pixels. The characters could also be redefined which wasn't common
at the time.
When combined with the most common graphics card one could display
both text and graphics simultaneously at the native 800x300 raster.
It had a 8MHz 8086 as standard and could be ordered with HDD from the
beginning (IIRC). It also had a real (though primitive SN74???) sound chip.

Actually still miss that system, nostalgia...

[Robert Wessel]

On Wed, 18 Oct 2017 09:50:45 +0200, Terje Mathisen
<terje.m...@tmsw.no> wrote:

>Robert Wessel wrote:
>> On Tue, 17 Oct 2017 16:38:03 -0400, paul wallich <p...@panix.com> wrote:
>>> FWIW, there were 8086-based machines in (low-volume) production years
>>> before the PC -- the 8088 was just there for cost reduction. So spinning
>>> an 8086 motherboard would have been easy and a simple way to higher
>>> performance (although compatibility might have been fun.
>>
>>
>> A number of folks built IBM PC compatibles with 8086, not least IBM
>> itself (some models of the PS/2 25 and 30 were 8086).
>>
>Are you sure?
>
>I would almost be willing to bet that the 80186 would have been the
>alternative to the 286 if they wanted a lower-end option at that point
>in time.

Yep. 8MHz 8086.

https://en.wikipedia.org/wiki/IBM_Personal_System/2#Models

[Megol]

On Wednesday, October 18, 2017 at 9:50:47 AM UTC+2, Terje Mathisen wrote:
> Robert Wessel wrote:
> > On Tue, 17 Oct 2017 16:38:03 -0400, paul wallich <p...@panix.com> wrote:
> >> FWIW, there were 8086-based machines in (low-volume) production years
> >> before the PC -- the 8088 was just there for cost reduction. So spinning
> >> an 8086 motherboard would have been easy and a simple way to higher
> >> performance (although compatibility might have been fun.
> >
> >
> > A number of folks built IBM PC compatibles with 8086, not least IBM
> > itself (some models of the PS/2 25 and 30 were 8086).
> >
> Are you sure?
>
> I would almost be willing to bet that the 80186 would have been the
> alternative to the 286 if they wanted a lower-end option at that point
> in time.

I'm sure - IBM had some 8086 machines in their PS/2 lineup. Wikipedia say
those were called Model 25.
The 80186 wasn't used much as it wasn't really compatible with the IBM
PC design and AFAIK it wasn't really possible to switch it to a compatible
mode (but could be wrong, didn't do any lower level coding on those).

Anyway using a 80186 but not using the integrated (non-PC compatible)
functionality seems like a worse choice than using the then cheap and
multi-sourced 8086. Unless the 8086 was soldered one could simply replace
it with a NEC V20 (V30?) and get more performance.

[Terje Mathisen]

I see you a Wikipedia article and raise you this:

"The Model 30 had either an 8086 or 286 CPU", right?

Well, the 186 was probably socket compatible with the 286, while the
8086 most definitely wasn't.

SW wise though, the 186 was effectively a working 8086: I.e it had fixed
the most glaring bugs in the 8086 like having an interrupt on top of an
instruction with two or more prefix bytes, but had absolutely none of
the 286 protected mode stuff.

I'm not too surprised that this could have been called an 86 vs a 286,
but unless they had two different "Model 30" machines with incompatible
cpu sockets, then I still think it likely that it was a 186.

Anyone with a motherboard photo?

[Robert Wessel]

On Wed, 18 Oct 2017 17:23:32 +0200, Terje Mathisen

<terje.m...@tmsw.no> wrote:

>Robert Wessel wrote:
>> On Wed, 18 Oct 2017 09:50:45 +0200, Terje Mathisen
>> <terje.m...@tmsw.no> wrote:
>>
>>> Robert Wessel wrote:
>>>> On Tue, 17 Oct 2017 16:38:03 -0400, paul wallich <p...@panix.com> wrote:
>>>>> FWIW, there were 8086-based machines in (low-volume) production years
>>>>> before the PC -- the 8088 was just there for cost reduction. So spinning
>>>>> an 8086 motherboard would have been easy and a simple way to higher
>>>>> performance (although compatibility might have been fun.
>>>>
>>>>
>>>> A number of folks built IBM PC compatibles with 8086, not least IBM
>>>> itself (some models of the PS/2 25 and 30 were 8086).
>>>>
>>> Are you sure?
>>>
>>> I would almost be willing to bet that the 80186 would have been the
>>> alternative to the 286 if they wanted a lower-end option at that point
>>> in time.
>>
>> Yep. 8MHz 8086.
>>
>> https://en.wikipedia.org/wiki/IBM_Personal_System/2#Models
>>
>I see you a Wikipedia article and raise you this:
>
>"The Model 30 had either an 8086 or 286 CPU", right?
>
>Well, the 186 was probably socket compatible with the 286, while the
>8086 most definitely wasn't.

It wasn't socket compatible. The 80186 had a bunch of standard
peripherals on-chip (as did the 80188), which used a bunch of the
pins. That was also the main reason it was rarely used: none of those
peripherals were actually PC compatible. You *could* use it, and a few
vendors did, but you had to disable all the on-board stuff, and
surround it with the "normal" PC compatible stuff.

http://datasheets.chipdb.org/AMD/8018x/80186/amd-80186.pdf
http://www.dmi.unict.it/~santoro/teaching/tfa/intel-80286.pdf

Amongst other things, the 186 had a multiplexed address databus like
the 8086, the 286 did not.

>SW wise though, the 186 was effectively a working 8086: I.e it had fixed
>the most glaring bugs in the 8086 like having an interrupt on top of an
>instruction with two or more prefix bytes, but had absolutely none of
>the 286 protected mode stuff.
>
>I'm not too surprised that this could have been called an 86 vs a 286,
>but unless they had two different "Model 30" machines with incompatible
>cpu sockets, then I still think it likely that it was a 186.

Different motherboards.

>Anyone with a motherboard photo?

There's a bunch out there, just half of them are clearly not of the
right system(s). Search for IBM 8525.

[Robert Swindells]

On Wed, 18 Oct 2017 17:23:32 +0200, Terje Mathisen wrote:

> Robert Wessel wrote:
>> On Wed, 18 Oct 2017 09:50:45 +0200, Terje Mathisen
>> <terje.m...@tmsw.no> wrote:
>>
>>> Robert Wessel wrote:
>>>> On Tue, 17 Oct 2017 16:38:03 -0400, paul wallich <p...@panix.com>
>>>> wrote:
>>>>> FWIW, there were 8086-based machines in (low-volume) production
>>>>> years before the PC -- the 8088 was just there for cost reduction.
>>>>> So spinning an 8086 motherboard would have been easy and a simple
>>>>> way to higher performance (although compatibility might have been
>>>>> fun.
>>>>
>>>>
>>>> A number of folks built IBM PC compatibles with 8086, not least IBM
>>>> itself (some models of the PS/2 25 and 30 were 8086).
>>>>
>>> Are you sure?
>>>
>>> I would almost be willing to bet that the 80186 would have been the
>>> alternative to the 286 if they wanted a lower-end option at that point
>>> in time.
>>
>> Yep. 8MHz 8086.
>>
>> https://en.wikipedia.org/wiki/IBM_Personal_System/2#Models
>>
> I see you a Wikipedia article and raise you this:
>
> "The Model 30 had either an 8086 or 286 CPU", right?

The next paragraph states "Later ISA PS/2 models comprised the Model
30-286 (a Model 30 with an Intel 286 CPU ..."

My memory is that this was introduced quite a while after the original
(8086) Model 30.

I had an ACT Apricot that used an 8086 (+8089), and later an Olivetti M24
that I upgraded from the 8086 to a NEC V30.

[timca...@aol.com]

The 186 used an 8086 system bus.

The 286 bus was different.

The Tandy 2000 computer used the 186.

I did some very basic benchmarks back then. An 8Mhz (no wait state) 186
was about 10% slower than a 6 Mhz (1 wait state) 286.

There were several "clones" that used the 8086 instead of the 8088, one
was built by AT&T, I think Eagle did another.

The NEC V20 (drop in replacement for the 8088) had execution times almost
identical to a 188, (same for V30 & 186) (and the NEC chip did most of the 186 new instructions). The NEC V20 had a couple of extra instructions (REPC, REPNC, and insert/extract bit fields), plus it could switch modes and execute 8080 instructions.

- Tim

[Terje Mathisen]

OK, I consider myself educated here, thanks!

I took over responsibility for all IBM-compatible PCs in Hydro in 1984.
At the time Hydro was the largest corporation in the country, with
subsidiaries/factories/offices in 100+ countries worldwide and more than
70K employees.

I tested a bunch of those "almost but not exactly" compatible machines ,
including the Apricot and that particular Olivetti model. :-)

The main test was of course MS Flight Simulator which broke all the
rules in the book by going directly to the HW pretty much everywhere.

For some of those machines the only real problem was that they
usedeither an 8086 or ran the 8088 at higher speed, either of those
meant that the hand-coded fixed timing loops ran too fast.

[MitchAlsup]

On Friday, October 13, 2017 at 8:06:41 PM UTC-5, George Neuner wrote:
> On Fri, 13 Oct 2017 15:02:41 -0700 (PDT), MitchAlsup
> <????????????@aol.com> wrote:
>
> >I had a 33 Mhz 486 machine, running windows 3.11, got tired of the the
> >constant swapping, and changed from 4 MB to 20MB. This was too much for
> >W 3.11 so I used the upper 4 MB as a disk cache and 16 MB as main memory.
> >
> >A few years later, my wife got a 66 MHz 586 with 8MB, and my lowly 486
> >would wipe the floor in comparison to her machine. Two things made the
> >lower performing machine run better, 486 had more main memory, and second
> >it had 2 disks, one serving as the swap partition. My workload was more
> >difficult than hers .....
> >
> >YMMV.
>
> FWIW:
>
> It was fairly well known back then that you could get a significant
> performance boost for ALU code on the P5 by compiling for the i486
> instead.

In my case, all of the code arrived precompiled as <relatively> std
applications like *.xls, *.doc,...

So, I was not given this option..........

[Chris]

On 10/18/17 00:19, Bruce Hoult wrote:

>
> Well, you know ... Kildall was out flying, or wouldn't sign an NDA for a meeting he didn't know the purpose of (or possibly even the identity of the participants). And the Microsoft guys wore suits, and promised they had something they didn't have.
>

Out flying ?, lol. I mean, who wants to transact business on
a good day for flying ?.

Fwir, Kildall was something of an iconoclast, nothing wrong
with that, but probably more interested in tech than running
a business. Understand that completely...

Regards,

Chris

[paul wallich]

The story that came out of DR at the time was that it was the NDA
(which, if it was written the way IBM's lawyers liked to write things
back then, would have led to potential liability for things disclosed at
the meeting even if DR already knew them, or learned them later from
other sources. A little like the rules for cleared people talking about
potentially classified material). My impression was the msft had little
or nothing to lose in the area.

(If that's true, it's yet another case of the old saw about successful
companies being hamstrung by concerns that a new product will
cannibalize existing product lines.)

paul

[Rob]

bitrex <bit...@de.lete.earthlink.net> wrote:
> The article here states regarding the 386SX at 16MHz
>
> "The 386 was a huge advance but you'd never know it from one of these
> little slugs — they were usually out-performed by the better 286s."
>
> "Even with 4MB or 8MB RAM, you wouldn't want to run Windows 3.1 on a
> 386SX-16 though. The SX-33s and DX-40s that followed soon after were
> vastly faster."

In the 286 and early 386 days, most users were still abusing the
processor to run 86 code (MS-DOS, Windows 3.1) which used only some
of the advanced features to provide memory bank switching (EMS, XMS).

However, the difference between 286 and 386 was that you finally could
have a demand-paged virtual memory system. They could run Linux.

With a job that suited that environment, a 386 certainly was better.
Running MS-DOS wasn't such a job.

[Bruce Hoult]

On Friday, October 20, 2017 at 5:06:10 PM UTC+3, Rob wrote:
> bitrex <bit...@de.lete.earthlink.net> wrote:
> > The article here states regarding the 386SX at 16MHz
> >
> > "The 386 was a huge advance but you'd never know it from one of these
> > little slugs — they were usually out-performed by the better 286s."
> >
> > "Even with 4MB or 8MB RAM, you wouldn't want to run Windows 3.1 on a
> > 386SX-16 though. The SX-33s and DX-40s that followed soon after were
> > vastly faster."
>
> In the 286 and early 386 days, most users were still abusing the
> processor to run 86 code (MS-DOS, Windows 3.1) which used only some
> of the advanced features to provide memory bank switching (EMS, XMS).
>
> However, the difference between 286 and 386 was that you finally could
> have a demand-paged virtual memory system. They could run Linux.

That's a bit disingenuous. Linux was of course specifically designed around the capabilities of the 386.

*Unix* (at least old versions) runs on machines without a demand-paged virtual memory system. In particular, it runs on the PDP-11.

The bigger and faster PDP-11's had an optional MMU called the KT11-C which provided 48 segment descriptors (base and length). The basic 64 KB address range was divided into 8 segments starting at each 8 KB virtual boundary. Each segment could have a length of any multiple of 64 bytes, from 0 to 8 KB. The start of each segment could be mapped into the 256 KB physical address space at any 64 byte boundary. Instruction fetch and data accesses use different segments (allowing 64 KB program plus 64 KB data), and kernel, supervisor, and user modes use different sets of segments.

Each segment had not-present, accessed and dirty bits so you could in fact treat it them as 8 KB demand paged virtual memory pages -- but you don't have many of them!

[Brian G. Lucas]

And Unix (in the form of PC/IX) ran on the IBM PC/XT. Getting the shell to
work required some changes.

[timca...@aol.com]

The 286 supported virtual memory just fine, it was segment based
instead of page based, but certaintly not the first computer to
do that.

Some examples of OS'es that ran and supported virtual
memory on the 286:
Coherent
OS/2
Various ports of AT&T Unix (includes PC/IX and Xenix).

- Tim

[already...@yahoo.com]

Rob never claimed otherwise.
He very specifically mentioned *demand-paged* virtual memory.

[Rick C. Hodgin]

On Friday, October 13, 2017 at 7:29:47 AM UTC-4, bitrex wrote:
> "The 386 was a huge advance but you'd never know it from one of these
> little slugs — they were usually out-performed by the better 286s."

The 80386 is not obsolete technology via its ISA, but only
in that it's in a very old process technology and core design.

The core itself was only 275K transistors. Today, 16 of
them could be arrayed, on modern process technologies, with
a notable bump in clock speed, with wide parallel throughput,
a large L1 and L2 cache, making it suitable for many apps.

The 80386 could also be extended to a greater pipeline depth
and go even faster with little difficulty.

People in hardware industries are so used to rapid tick tock
cycles for money goals, they don't stop to realize how truly
useful some things could be due to low power requirements
on modern process technologies, without changing anything
else.

How fast and small would a 10nm 80386 be with only 275K
transistors? That's a question I'd like to see answered.
Add in 16 or 32 cores per die, L1, L2, faster bus, and I think
it would be a winner becauser so many existing tools and
software libraries are standing by.

Thank you,
Rick C. Hodgin

[John Dallman]

In article <slrnouk0lb...@xs9.xs4all.nl>, nom...@example.com (Rob)
wrote:

> With a job that suited that environment, a 386 certainly was better.
> Running MS-DOS wasn't such a job.

It was if you had software that could use EMS or XMS. Those worked much
more smoothly when implemented by a good 386 memory manager (QEMM was the
best, but HIMEM.SYS and EMM386.SYS were OK) than on a 286 or lesser
processor with expansion cards.

John