?

Log in

No account? Create an account

Sat, Jan. 30th, 2016, 07:37 pm
Fallen giants - comparing the '80s second-generation home computers

A friend of mine who is a Commodore enthusiast commented that if the company had handled it better, the Amiga would have killed the Apple Mac off.

But I wonder. I mean, the $10K Lisa ('83) and the $2.5K Mac ('84) may only have been a year or two before the $1.3K Amiga 1000 ('85), but in those years, chip prices were plummeting -- maybe rapidly enough to account for the discrepancy.

The 256kB Amiga 1000 was half the price of the original 128kB Mac a year earlier.

Could Tramiel's Commodore have sold Macs at a profit for much less? I'm not sure. Later, yes, but then, Mac prices fell, and anyway, Apple has long been a premium-products-only sort of company. But the R&D process behind the Lisa & the Mac was long, complex & expensive. (Yes, true, it was behind the Amiga chipset, too, but less so on the OS -- the original CAOS got axed, remember. The TRIPOS thing was a last-minute stand-in, as was Arthur/RISC OS on the Acorn Archimedes.)

The existence of the Amiga also pushed development of the Mac II, the first colour model. (Although I think it probably more directly prompted the Apple ][GS.)

It's much easier to copy something that someone else has already done. Without the precedent of the Lisa, the Mac would have been a much more limited 8-bit machine with a 6809. Without the precedent of the Mac, the Amiga would have been a games console.


I think the contrast between the Atari ST and the Sinclair QL, in terms of business decisions, product focus and so on, is more instructive.
The QL could have been one of the imporant 2nd-generation home computers. It was launched a couple of weeks before the Mac.
But Sinclair went too far with its hallmark cost-cutting on the project, and the launch date was too ambitious. The result was a 16-bit machine that was barely more capable than an 8-bit one from the previous generation. Most of the later 8-bit machines had better graphics and sound; some (Memotech, Elan Enterprise) as much RAM, and some (e.g. the SAM Coupé) also supported built-in mass storage.
But Sinclair's OS, QDOS, was impressive. An excellent BASIC, front & centre like an 8-bit machine, but also full multitasking, modularity so it readily handled new peripherals -- but no GUI by default.
The Mac, similarly RAM deprived and with even poorer graphics, blew it away. Also, with the Lisa and the Mac, Apple had spotted that the future lay in GUIs, which Sinclair had missed -- the QL didn't get its "pointer environment" until later, and when it did, it was primitive-looking. Even the modern version is:



Atari, entering the game a year or so later, had a much better idea where to spend the money. The ST was an excellent demonstration of cost-cutting. Unlike the bespoke custom chipsets of the Mac and the Amiga, or Sinclair's manic focus on cheapness, Atari took off-the-shelf hardware and off-the-shelf software and assembled something that was good enough. A decent GUI, an OS that worked well in 512kB, graphics and sound that were good enough. Marginally faster CPU than an Amiga, and a floppy format interchangeable with PCs.
Yes, the Amiga was a better machine in almost every way, but the ST was good enough, and at first, significantly cheaper. Commodore had to cost-trim the Amiga to match, and the first result, the Amiga 500, was a good games machine but too compromised for much else.

The QL was built down to a price, and suffered for it. Later replacement motherboards and third-party clones such as the Thor fixed much of this, but it was no match for the GUI-based machines.

The Mac was in some ways a sort of cut-down Lisa, trying to get that ten-thousand-dollar machine down to a more affordable quarter of the price. Sadly, this meant losing the hard disk and the innovative multitasking OS, which were added back later in compromised form -- the latter cursed the classic MacOS until it was replaced with Mac OS X at the turn of the century.

The Amiga was a no-compromise games machine, later cleverly shoehorned into the role of a very capable multimedia GUI coomputer.

The ST was also built down to a price, but learned from the lessons of the Mac. Its spec wasn't as good as the Amiga, its OS wasn't as elegant as the Mac, but it was good enough.

The result was that games developers aimed at both, limiting the quality of Amiga games to the capabilities of the ST. The Amiga wasn't differentiated enough -- yes, Commodore did high-end three-box versions, but the basic machines remained too low-spec. The third-generation Amiga 1200 had a faster 68020 chip which the OS didn't really utilise, it had provision for a built-in hard disk which was an optional extra. AmigaOS was a pain to use with only floppies, like the Mac -- whereas the ST's ROM-based OS was fairly usable with a single drive. A dual-floppy-drive Amiga was the minimum usable spec, really, and it benefited hugely from a hard disk -- but Commodore didn't fit one.

The ST killed the Amiga, in effect. By providing an experience that was nearly as good in the important, visible ways, Commodore had to price-cut the Amiga to keep it competitive, hobbling the lower-end models. And as games were written to be portable between them both without too much work, they mostly didn't exploit the Amiga's superior abilities.

Acorn went its own way with the Archimedes -- it shared almost no apps or games with the mainstream machines, and while its OS is still around, it hasn't kept up with the times and is mainly a curiosity. Acorn kept its machines a bit higher-end, having affordable three-box models with hard disks right from the start, and focused on the educational niche where it was strong.

But Acorn's decision to go its own way was entirely vindicated -- its ARM chip is now the world's best-selling CPU. Both Microsoft and Apple OSes run on ARMs now. In a way, it won.

The poor Sinclair QL, of course, failed in the market and Amstrad killed it off when it was still young. But even so, it inspired a whole line of successors -- the CST Thor, the ICL One-Per-Desk (AKA Merlin Tonto, AKA Telecom Australia ComputerPhone), the Qubbesoft Aurora replacement main board and later the Q40 and Q60 QL-compatible PC-style motherboards. It had the first ever multitasking OS for a home computer, QDOS, which evolved into SMSQ/e and moved over to the ST platform instead. It's now open source, too.

And Linus Torvalds owned a QL, giving him a taste for multitasking so that he wrote his own multitasking OS when he got a PC. That, of course, was Linux.

The Amiga OS is still limping along, now running on a CPU line -- PowerPC -- that is also all but dead. The open-source version, AROS, is working on an ARM port, which might make it slightly more relevant, but it's hard to see a future or purpose for the two PowerPC versions, MorphOS and AmigaOS 4.

The ST OS also evolved, into a rich multitasking app environment for PCs and Macs (MagiC) and into a rich multitasking FOSS version, AFROS, running on an emulator on the PC, Aranym. A great and very clever little project but which went nowhere, as did PC GEM, sadly.

All of these clever OSes -- AROS, AFROS, QDOS AKA SMSQ/E. All went FOSS too late and are forgotten. Me, I'd love Raspberry Pi versions of any and all of them to play with!

In its final death throes, a flailing Atari even embraced the Transputer. The Atari ABAQ could run Parhelion's HELIOS, another interesting long-dead OS. Acorn's machines ran one of the most amazing OSes I've ever seen, TAOS, which nearly became the next-generation Amiga OS. That could have shaken up the industry -- it was truly radical.

And in a funny little side-note, the next next-gen Amiga OS after TAOS was to be QNX. It didn't happen, but QNX added a GUI and rich multimedia support to its embedded microkernel OS for the deal. That OS is now what powers my Blackberry Passport smartphone. Blackberry 10 is now all but dead -- Blackberry has conceded the inevitable and gone Android -- but BB10 is a beautiful piece of work, way better than its rivals.

But all the successful machines that sold well? The ST and Amiga lines are effectively dead. The Motorola 68K processor line they used is all but dead, too. So is its successor, PowerPC.

So it's the two niche machines that left the real legacy. In a way, Sinclair Research did have the right idea after all -- but prematurely. It thought that the justification for 16-bit home/business computers was multitasking. In the end, it was, but only in the later 32-bit era: the defining characteristic of the 16-bit era was bringing the GUI to the masses. True robust multitasking for all followed later. Sinclair picked the wrong feature to emphasise -- even though the QL post-dated the Apple Lisa, so the writing was there on the wall for all to see.

But in the end, the QL inspired Linux and the Archimedes gave us the ARM chip, the most successful RISC chip ever and the one that could still conceivably drive the last great CISC architecture, x86, into extinction.

Funny how things turn out.

Sat, Jan. 30th, 2016 10:02 pm (UTC)
waistcoatmark

Did the ST really have an OS? It could copy files and launch programs, but it was about as much an Operating System as DOS was. Granted I never programmed it, so it might have had some elegant backend hidden behind it's limited GEM exterior, but if so, it was *very* well hidden...

Sat, Jan. 30th, 2016 11:55 pm (UTC)
uon

What counts as "really an OS"? It had what counted at the time, which was basically a file system, some thin hardware abstraction, and a couple of utility functions. The 68000 had no proper MMU, so a lot of what you'd expect from a modern OS was regarded as impossible. (It wasn't actually impossible, but you had to resort to bizarre tricks to make it work, like using two CPUs, which wasn't really viable for a home computer then).

There was almost Proper Unix. Oh, if only...

Sun, Jan. 31st, 2016 11:59 am (UTC)
pndc

An operating system doesn't *have* to sandbox processes from each other. It just has to ensure that processes play nicely and can share resources.

IMO, one of the important things to share is control, and AmigaOS was the only widely-available microcomputer OS that had proper pre-emptive multitasking. RiscOS, Atari TOS and MacOS Classic were single-tasking systems which that had co-operative multitasking bodged on later as an afterthought, and it shows. You can't safely preempt a program that believes it has full control of the machine.

(Yes, AmigaOS contained loads of flaws that made its multitasking much less effective than it could have been. It was still less awful than the others.)

A Unix process is a specialised virtual machine which believes it has full control, and preemption switches in a completely different virtual machine. This is why Unix excels at multiprocessing, but threading and signals within a virtual machine are a bit of a disaster.

Sun, Jan. 31st, 2016 01:34 pm (UTC)
liam_on_linux

I love that story. Not sure if you originally pointed me at it. I don't understand all the implementation detail but it sounds like a cool hack.

DR's CP/M family definitely grew into "proper OSes" on other hardware. A descendant of Concurrent CP/M called FlexOS survived until quite recently on IBM point of sale kit. It even had a multitasking GUI, X-GEM.

Mon, Feb. 1st, 2016 06:37 pm (UTC)
liam_on_linux

The last living descendant of DR's multitasking OSes, FlexOS, became IBM 4690:

https://en.wikipedia.org/wiki/4690_Operating_System

This is still supported and is now sold by Toshiba:

http://www-03.ibm.com/products/retail/products/software/4690/index.html

Tue, Feb. 2nd, 2016 07:25 pm (UTC)
waistcoatmark

My first job involved programming 4680 POS systems...

Mon, Feb. 1st, 2016 06:09 pm (UTC)
liam_on_linux

Oh, definitely, yes.

Digital Research wrote CP/M, the original cross-platform OS for micros. Later, it grew into CP/M-86, the PC version, and then Concurrent CP/M, a multitasking/multiuser version. This had DOS compatibility grafted on -- since MS-DOS 1.0 was a clone of CP/M-86 anyway -- and became Concurrent DOS, a multiuser/multitasking DOS-compatible OS. The 386 version was very capable, but LANs killed it off.

So DR moved the multitasking out into a separate module and made CDOS into DR-DOS, a single-tasking MS-DOS replacement. The first version was called v3.41 to signify that it was broadly comparable to MS-DOS 3.3 but slightly better -- notably it had big-disk support (i.e. >32MB partitions) which only Compaq DOS 3.31 offered in the MS family.

It later developed into DR-DOS 5 (added memory management), DR-DOS 6 (added whole-disk compression) and Novell DOS 7 (added peer-to-peer networking). That's now FOSS: http://www.drdosprojects.de/

Meanwhile, the ST's kernel was GEMDOS, sort of the DOS-like functionality of DR-DOS ~3.41 but built from the CP/M-68K codebase, the version for 68000 chips.

The whole command-line shell and all the DOS commands were omitted; instead it booted GEM directly.

PC GEM was crippled because Apple won a look-&-feel lawsuit. Later, it was open-sourced and the missing overlapping windows etc. were put back in: http://www.owenrudge.net/GEM/

The GUI is part of FreeDOS today: http://www.freedos.org/

But the Atari version was never crippled.

Since it was modular, people re-implemented bits. Some did it as commercial/closed source -- that hangs on as MagiC: http://www.application-systems.de/magicpc/

Others did it, piecemeal, as FOSS. That evolved into FreeMINT: http://wiki.sparemint.org/index.php/FreeMiNT

And then into AFROS, a complete stack from firmware up, which runs on its own emulator, ARANYM: http://aranym.org/

So, yes, ultimately, a Unix-like multitasking OS built on a DOS-like kernel and GEM.


Edited at 2016-02-01 07:04 pm (UTC)

Mon, Feb. 1st, 2016 07:07 pm (UTC)
liam_on_linux

Incidentally, Real/32, the last descendant of CDOS, is still on sale!

http://www.imsltd.com/

The company tried to develop a Linux-based successor but it went nowhere.

Sun, Jan. 31st, 2016 12:23 am (UTC)
vicarage

You don't mention the machine I bought in '84 an Amstrad CPC, or the serious minded PCW. Weren't they, with their excellent value, the truely sucessful home machines of their period, in the uk at least?

Sun, Jan. 31st, 2016 04:57 pm (UTC)
(Anonymous)

They were both "first gen" (if you count amiga/st as second gen, as per the title). 8-bit, resorting to hacks even to access 128k of ram etc.

Mon, Feb. 1st, 2016 07:01 pm (UTC)
liam_on_linux

Exactly.

The thing with the 8-bitters is that there were so many of them, it's tricky to distinguish clear influences.

One of the best-designed was the BBC Micro. Very capable and elegant and clean. It bequeathed a bit of genetics to the Archimedes, and that's the healthiest product line by far today.

CBM bought the Amiga in, so it has no commonality with the VIC20/C64. Actually it was built by Atari designers who did the Atari 400/800.

The ST shares no kinship with its Atari forebears -- Commodore hired 'em.

The Commodore OS is alive: http://www.amigaos.net/

You can buy at least 2 models of brand-new Amiga today:

AmigaOne 500:

http://www.acube-systems.biz/index.php?page=hardware&pid=7

AmigaOne X1000:

http://www.a-eon.com/

The 2 unrelated Sinclair lines died off, except in the Eastern Bloc, where the Spectrum flourished & developed into 16-bit machines with ISA slots and hard disks (!).

Most of the smaller players went nowhere.

Sun, Jan. 31st, 2016 09:19 pm (UTC)
waistcoatmark

Going by Liam's definition of St/Amiga as 2nd generation, then those were definitely 1st gen: 8-bit processors, unable to access more than 64k of memory without having to jump through nasty hoops.

Mon, Feb. 1st, 2016 06:51 pm (UTC)
liam_on_linux

A fun question is: was Acorn's Archimedes range 2nd-gen (because it was contemporaneous with the Amiga & ST) or 3rd-gen (because it's 32-bit, had a good enough CPU to compete effectively with x86, as it still does today.

OK, there are (almost) no ARM desktops today. Almost:

A9home: http://www.advantage6.com/products/A9home.html

The ARMini (a Beagleboard XM in a case): http://www.armini.co.uk/

I reviewed it:

http://www.theregister.co.uk/2012/05/10/product_round_up_arm_mini_computers_the_best_and_the_rest/?page=2

And a new one coming, Titanium:

https://www.riscosopen.org/news/articles/2015/10/23/preview-of-a-whole-new-risc-os-platform

I favour the view that the Miggy & ST were the last gasp of the 68K, actually a 1970s chip. OK, yes, faster versions came, as with the Mac -- but Apple jumped ship to PowerPC and Atari and Commodore didn't and died. Whereas the Archie was far ahead of its time.

This remains possibly my single favourite computer-product review of all time:

http://chrisacorns.computinghistory.org.uk/docs/Mags/PCW/PCW_Aug87_Archimedes.pdf

Worth a read. Lines like:

"It loads huge programs with a faint burping noise, in the time it takes to blink an eye."

The joys of an 8Mz ARM2 running an ST-506 hard disk with zero interleave -- sucking sectors off the disk as fast as the controller could shift them. Glorious.

Mon, Feb. 1st, 2016 06:23 pm (UTC)
liam_on_linux

Well, they were 8-bits. The lines between and within generations are kind of blurry, but the first tranche of the first gen were simple monochrome text-only 8-bits -- often build-it-yourself. KIM, UK101, CBM PET, Apple 1. Often 4-8kB RAM.

The second slice of G1 was the early ready-made 8-bitters, many but not all with colour: Apple ][, VIC-20, Atari 400/800, ZX-81. Small amounts of ram -- under 64kB.

Then the next batch: beginning of the '80s. Oric-1, Spectrum, BBC Micro, C64, Memotech, MSX. Often had a full memory map, 64kB: some ROM and the rest RAM. Many didn't have a floppy disk system as standard, so there were competing standards, possibly incompatible.

Then the next, in the mid-'80s: the Amstrad CPCs, the Elan Enterprise, Oric Atmos, BBC Master, MSX2. Better specs but still built down to a price somewhat. Often offered >64kB RAM via bank-switching and offered or even came with floppy drives.

The last gasp of the 8-bits was the final super-8-bit machines. Not so successful as by now the 16-bitters were around. MSX2+ and MSX3, and my favourite, the SAM Coupé. Arguably the Amstrad PCW, a business machine: no sound, no colour, hi-res graphics & a vast amount of RAM for a CP/M machine -- 256kB or even 512kB, meaning practical single-drive operation with the OS and apps in a permanently-available-as-standard RAMdrive.

I like the late-era machines, which were often very uncompromised -- but frankly, a next-gen 16-bit machine such as an ST or an Amiga was a better buy. Some had ST-level graphics but not the CPU grunt to throw around large colour bitmaps. You couldn't use >64kB at once so no fancy programming environments, just BASIC with some extensions. Nice, but ultimately a bit pointless: flogging a dead horse.

Edited at 2016-02-01 06:23 pm (UTC)

Sun, Jan. 31st, 2016 11:32 am (UTC)
pndc

The first Mac nearly shipped with the same awful 8 bit wide 68008 CPU that dogged the QL. It was attractive to Uncle Clive because it was cheaper, and to Apple team for much the same reason. Apple selected it because it meant that they could use a single 8 bit wide ROM and RAM chip. Back then (and to an extent even now) adding more pins to a chip makes it disproportionately more expensive, so reducing pin count was (is) key and if you wanted a 16 bit wide bus, you put two 8 bit chips in parallel. The 68000 itself was an example of this disproportional expense with its profligate 64 pins. The 68008 had 48 pins which was also pushing it. Fortunately(!), Apple's developers couldn't keep code bloat under control and the firmware grew larger than the a single ROM, at which point redesigning the machine to use the 16 bit wide 68000 was pretty much inevitable. So the Macintosh dodged that bullet otherwise it might have been just a footnote in computer history.

(Playing silly buggers with restricted bus widths to get away with a single memory chip is also why the Acorn Electron was such a dog.)

Amiga absolutely did die due to management incompetence at head office. To a rough approximation, nobody bought it in the USA because there was chronic underinvestment in both marketing and product development. Meanwhile, it was flying off the shelves in Europe, and there was a lot of frustration in the Commodore UK management team at the hippies across the Pond. If David Pleasance et al had been running the American operation, Apple may again have been just a footnote :) Commodore UK went into *voluntary* liquidation when head office went pop. They were perfectly solvent, but no longer served any useful purpose as they didn't own the IP and couldn't continue making and selling machines.

The 68000 itself is dead, but the instruction set lives on. Freescale (a Motorola spin-off) noticed that compilers mostly used the more RISCy subset of the instructions, so they stripped out all of the difficult CISC bits to make it go much faster. The result was Coldfire, which is still sold and used in embedded systems. Freescale was recently sold to NXP (a Philips spin-off) in the Netherlands. Philips have history with m68k and made a "68070" in the past which ended up in its ill-fated CDi.

Stripping back the m68k ISA actually started as early as 1991 with the 68040. The 68020 was an example of second-system effect in that a whole load of extremely complex CISC instructions and addressing modes were added, even though some of them had dubious utility. The corresponding 68881 FPU and 68851 MMU were also rather complex beasts. The MMU was merged into the 68030, and the FPU joined it in the 68040. Because the 68040 blew its transistor budget, it didn't support all of the 68881 instructions and the MMU was completely replaced with a much simpler design. The 68060 also dropped some of the 68020 instructions. It wasn't until Coldfire that they also dropped 68000 instructions and was no longer binary compatible with any of the previous CPUs.

Sun, Jan. 31st, 2016 11:57 pm (UTC)
uon

To be fair, I remember reading dark hints in the documentation at the time that some of the weird stuff in the 68020 (the CALLM/RTM opcodes in particular) were only there because one particular hardware vendor leaned very heavily on Motorola to get them in there, and they were stripped out in the 68030 and subsequent chips.

Mon, Feb. 1st, 2016 06:33 pm (UTC)
liam_on_linux

Good reply!

From my reading, as I understand it, the Mac started out based around the 6809. The 68000 was a later upgrade. I've not heard of a 68008 stage but I'm not saying you're wrong.

The evolution of the 680x0 series did seem even more undirected than that of the 32-bit x86s. Intel bolted stuff in as and when it could, then occasionally realised it had overshot and released a cut-down (e.g. 386SX) or intentionally crippled (e.g. 486SX) model.

Motorola, on the other hand, flailed.

But yes, Coldfire is still around, you're right, and there is an ST clone built around it, the FireBee:

http://acp.atari.org/

But the FPGA approaches seem to be winning IMHO -- some I never thought would make it have actually shipped!

MiniMig, an Original ChipSet Amiga clone:

http://amigakit.leamancomputing.com/catalog/product_info.php?products_id=777

MiST, originally an ST, now does Amiga & some 8-bitters:

http://www.lotharek.pl/product.php?pid=96

(Yes, I vaguely want one of all of them. No, I'm not going to.)

Tue, Feb. 2nd, 2016 11:23 am (UTC)
pndc

I slighly misremembered the story I read in Revolution in the Valley, which also turns out to be at http://www.folklore.org/StoryView.py?story=Five_Different_Macs.txt

Rather than selecting the 68008 directly, Burrell designed glue logic for the 68000 that hobbled it down to the same spec as the 68008 to produce the same cost-reduction on memory chips. This work was done in December 1980, whereas the 68008 was released in 1982. It seems likely that Apple would have chosen the 68008 were it available at the time.

Those various emulators are cute, but do rather look like reference designed bodged into cheap Maplin cases, rather than some nice retro hardware that also looks the part.

Sun, Jan. 31st, 2016 12:02 pm (UTC)
pndc

Also, if you haven't yet discovered http://www.filfre.net/, now is an excellent time to cancel all your appointments for the rest of the day, fetch a case of beer, and have a very long and enjoyable read about the early history of microcomputers.

Wed, Feb. 3rd, 2016 02:07 am (UTC)
(Anonymous): some addendum

Hi Liam,

There are a couple of thinks to add here:
-With Tramiel leaving Commodore, Commodore lost most of its engineers, Tramiel took his money also.
-So Commodore bought Amiga, which was a costly business. At the same time 8-bit sales tanked. Tramiel's business tactics(low C64 price) also meant low margin for Commodore.
-Tramiel also did not have much money, so he had to hurry to make some product available.
-So both companies were restricted by their low capital by the end of 1985.
-Amiga was not really that useful when released. It did not have proper HDD support. It did not have proper flickerfree high resolution mode. It had very small amount of RAM. 256k for the multitasing OS was nothing. Basically the first thing to do for the Amiga was to buy RAM expansion.
-Atari TOS1.0 was on floppy, just like with Amiga. But with the ST1040, it was in ROM. The Atari with 1Meg of Ram was cheaper than the Amiga.
-Apple had laser writer and Microsoft Write in 1985. Atari also got good business support because it had an awesome mono mode, which the Amiga hadn't until ECS added productivity mode.
-The Amiga1000 was too expensive for the home market. The Ataris did not kill the Amiga. After the A2000/A500 was introduced those were quite good sellers. So much so, that Commodore did quite well in 1989/1990/1991 financially. If something killed the Amiga it was the cheap 286@16MHz CPUs which took over the lowend segment in 1990/1991
-i could tell more but it is late here... good night.