Obsolete processors, 286 vs. 386

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.

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...

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.

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

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.

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

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.

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

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

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

George

On a sunny day (13 Oct 2017 21:12:37 GMT) it happened Jasen Betts wrote in :

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.

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.

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

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

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

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

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.

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

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.

"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)

Am 15.10.2017 um 17:27 schrieb Robert Wessel:

completely correct.

regards, Gerhard

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.

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

George

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.

6s."

Linux was still about 5 years in the future when the 386 was popular.

But when Linux appeared, most PC users were STILL running 86 code on their 386!! There were alternatives but they were hardly in use.

Remember it took 3 more years after Linux appeared before Windows95 changed that.