Re: Problems in Modernizing the IIgs, con't

[pubpc1@library.ucla.edu]

B.J. Major (B.J. and the Bear?) wrote,
Lewin Edwards wrote,
in response to my reply:
>>I talked to an friend of mine that uses and programs an Amiga 500
>>(a computer that uses the 68000) and asked him about these supposed
>>problems in upgrading the GS.  Here's what he has to say:
>
>Let me preface my comments here by saying that I am very familiar with
>the Amiga from a developer perspective, having performed a large number
>of hardware and software hacks on the revs 5 and 6a A500, so if you were
>hoping to choose an example about which I knew nothing in order to
>perform a little smoke-and-mirrors magic, you've been unlucky.

I'm insulted that you would even think that.  The very reason I
talked to an Amiga 500 user was because he was familiar with
the 68000 and you are familiar with the same chip.

>>Some Mac users have spammed csa2 and claim that the GS could never
>>be upgraded.
>
>Whilst I can't comment on what other people have been posting to csa2,
>since this message is at least partly directed at me, I feel it
>necessary to point out that I did not say the GS could never be
>upgraded, the original thread was about someone thinking of spending
>money realizing the 65832 design, and I pointed out some of the
>technical problems and mentioned that it wasn't really worth the effort,
>except from the purely hackerish point of view that "it would be cool to
>do". Which is, of course, a legitimate reason to do something. The
>argument has now moved on to whether or not it makes sense to develop
>the 65xxx series MPUs beyond their current state.

The reason the 65C816 32-bit derivative was never realized was
because Apple killed the GS.  Laser, a clone maker, was also
readying an '816 computer and others were to follow, but Apple
made it clear there was no support.  That's why WDC decided to
spend time and money on the microcontroller market instead.

To say it made no sense to develop the '816 beyond is current
state is like saying there was no sense to develop the original
68K into an '040.  The Pentium was derived from an 8080, so it
could be done.  There are many factors in a CPU chip besides the
data bus.  The 65C816 already has 24-bit addressing and in this
respect is similar to the 68000 and 80286 (all three used in
computers from '86).  The '286, using a memory addessing scheme
similar to the '816 (segmented memory), has evolved far beyond
this.  The '816 could have also.  It would have been damn
feasable!  The GS outsold the entire Mac line during its first
year.  The only reason sales dropped later was that Apple
deliberately cut advertising and marketing.

The folks that designed the Amiga's graphics chipset originally
approached Apple to put it into the II, but Apple refused, btw.
Such a chipset and an Ensoniq wavetable synthesizer in a single
computer with the Apple Interface would have been a killer sight!

>>They doubt the derived 65C816 32-bit chips could have been made.
>
>I never said that either. What I said was that it would have been
>commercially stupid for Apple to go on developing the 65xxx series MPUs,
>it would be stupid for them to do so now, and that I regard the idea of
>a hot-rodded IIgs as being fun and cool in a theoretical sense, but of
>no commercial value whatsoever.

It would have been SMART for Apple to continue developing the
65xxx series.  They already had a "full house" of customer base.
Most of us deserted Apple after it deserted us.  Look at what
bad shape Apple is in today.

Apple also did not have to wait for a 32-bit chip to develop the
GS.  Simply using a faster 20MHz 65C816 and a motherboard running
at that speed, plus Amiga type graphics coprocessors and more
modern versions of the Ensoniq sound chip would have been a big
improvement and renewed the GS in 1989-1990 (instead, Apple gave
us the ROM 3, the exact same GS except with more memory and more
ROM).

>>"That figures, we've been trying to convice them that they don't
>>have pre-emtive multitasking for ages and they don't get it.
>
>Ah, someone having a nice hearty bitch. So what about PEM? Your Amiga
>500 example lacks an MMU and doesn't have virtual memory or protected
>memory areas either. And the preemptive multitasking is not a hardware
>feature anyway, so it's completely irrelevant to the discussion. In
>AmigaDOS, the multitasking is driven by the vertical retrace interrupt.
>The Mac 512K and upwards have a vertical retrace interrupt service which
>could similarly be used to implement a fully preemptively multitasking
>OS, if someone could be bothered.

Then why did "someone" never implement preemtive multitasking in
MacOS?  Even the GS has it for non-desktop programs under GNO/ME.

As for the Amiga, be reasonable.  We're comparing the A500 to an
68000 Mac.  The Amiga actually did evolve to use later 68K
derivatives and these later models of the Amiga had an MMU.

>>"Clock for clock?  No way is the 68000 more efficient at all
>>tasks.  At the equivalent frequency, the 65C816 will beat a
>>68000 in many tasks.
>
>In artificial tests of non-real-programming-task instruction mixes only.

BS!  An GS accelerated to 7MHz will beat out a 7MHz black and
white 68000 based Mac in practically every damn task (word
processing, page layout, etc.) even though the GS is still
bottlenecked by the motherboard.  If the motherboard was also
running at 7MHz, it would fry the Mac alive.

>>The Amiga deals with this by using custom hardware graphics
>>chips, which is why our graphics are so much better than the
>>equivalent Macs and don't slow down the computer."
>
>That indicates a serious lack of understanding of the Amiga's
>architecture, too. The baseline A500 does not outperform any color
>Macintosh in terms of graphic performance because "the custom chips
>don't slow down the computer".

So the chips and the 68000 fight for time.  So what?  The end
result is that the custom graphics coprocessor moves graphics
faster than the 68000 ever could.  The 68000 could go to sleep
for all the user cares, as long as the job gets done faster.
That's why an A500 outperforms any 68000 Mac in graphics.

There were no color 68000 Macs.  The 68000 A500 beats out any
68000 Mac in graphics and it does so in color (more data moved)
while the Mac does so in 1-bit b&w (which should give it the
advantage).

>The reason that the blitter in Agnus
>makes animation tasks smoother and faster on the Amiga is because this
>device is essentially a custom MPU optimized for graphics operations.

Great, isn't that the point?

>Buss contention slows the 68000 to a crawl when you're using
>high-resolution graphics on the stock Amiga 500. Audio, disk and blitter
>DMA slow it down even more.

SO WHAT?  The 68K could go to sleep for all the user cares,
as long as the rest of the computer can get things done faster!

>Basically, doing a complex series of graphic ops on the Amiga involves
>setting up a list of blitter ops to be performed, installing a blitter
>DMA finished ISR which will fetch the next item pending from the list,
>and starting the first op manually. While that is running, you have a
>*little* CPU timeslice in the foreground, but that's really borrowed
>time. Agnus is chewing most of the RAM bandwidth.

You are saying the coprocessors hinder the CPU, but you neglect
the fact that during the time they take over from the CPU, they
can get their respective jobs done faster than the 68000 ever
could alone.  That's the point.  He was saying that you could
use a similar principle to build an 'Amiga-like' GS with a 20MHz
CPU and a 20MHz motherboard.  Here's what he said:

------------------
How would you upgrade a GS without a 32-bit 65C816 derivative?

"I would put the fastest 65C816 that WDC has (20MHz) into a
motherboard that runs at this speed.  Then I would use 'the
Amiga Principle,' using custom chips to off-load work from
the CPU.  Put the equivalent of the Amiga graphics chipset
in (that TurboRez card that was cancelled is a good equivalent)
that provides multiple bitplanes and a hardware blitter.
Give SVGA output, like that Second Sight card offers.  Put
in a 68882 math coprocessor unit like the one that was
available for the GS on a card.  Then round it off by putting
in a modern version of the GS's Ensoniq sound chip.  8-bit
Apple II emulation could be moved into software easily.  Use
an IDE hard drives for the biggest bang for the buck and put
in at least 4MB, expandable to 16MB."

Can this new GS evolve into something even better?

"Definitely yes.  If successful, Apple should move to upgrade
the GS's OS.  The first thing I would do is to slowly make the
GS toolbox reentrant.  I assume this is why you can preemtive
multitask under GNO/ME, but only have one Desktop program being
multitasked.  Then they could eventually migrate the platform
to PowerPC if things work out.  Keep the hardware graphics and
sound, port the GS toolbox to native PPC code, and emulate the
65C816."

The Mac folks say that this is like the PowerMac.

"Sure, except that it has sound and graphics features that we
Amiga and GS users like and it has the GS toolbox in native
code.  That's a big advantage.  Also, all you Apple II folks
would have such a GS now and Apple would be in much better
shape."

They say at some time backward compatibility must be cut

"Yes, but if you give users an upgrade path with backward
compatibility at each step, they will follow.  If you break
backward compatibility in a single step, they [the users]
will defect.  The GS is compatible with the IIe.  The
'Amiga-like' GS I am suggesting will break IIe compatibilty
but remain truly GS compatible.  Then the RISC based GS
will emulate the GS in software and eventually, after a
few years, go native.  At each step, give a few years in
between for users to migrate and be backward compatible
not with the very oldest model, but with at least the previous
one."

---------------

Now I understand, Lewin, that you are saying the coprocessor
chips may interfere with the 68000's processing time, but
with the coprocessors doing the job faster, who cares?
You may get a hit when multitasking, but at least AmigaDOS
can do preemptive multitasking.

Based on my conversation with him, I'd say Apple would have
had a winner using the following timetable:

GS w/1MB:  1986
GS/30 (7MHz processor/30MB hard drive): 1987
'Amiga-like' GS:  1990
PowerGS (65K emulation):  1994
PowerGS (all native):  1997

>>by Gateway 2000, so it can be done.  The market for computers
>>priced at $600 is large and left by the void formerly filled
>
>You can't be serious. GW2K is into interactive CATV. The personal
>computer market as personal computers is driven essentially by
>availability of Microsoft Office.
>
>If this fantasy was even remotely feasible, then Be would still be
>making BeBoxen.

You did not quote the full text above.  The point made was that
there is a market for an appliance computer.  Think about this:
Brother, Smith Corona, and Canon make $400 devices they call
word processors, but are in actuality PC/XT class computers with
VGA screens running GeoWorks.  I'm serious, these are in Office
Depot, Staples, and other office supply stores.  You say that
GW2K is reviving Amiga technology for CATV.  Great, that's
"appliance" technology.  Remember that most people still do not
have computers.  The Macintosh has failed as "the computer for
the rest of us."  Non-computer users don't trust it anymore than
the PC.  Similarly, capable PC systems under $K no longer exist.
You need something like that to run something like Microsoft
Office.

A 20MHz GS appliance computer bundled with key apps such as
AppleWorks GS, WordPerfect GS, SuperCalc 3a, Animasia, TCP/IP,
Spectrum and the SIS web browser, etc. and integrated with a
600dpi ink-jet printer body for $600 would sell.  If GeoWorks
word processors sell, so can GS appliance computers.  Also
consider that such a GS can be portable.  The 65C816 is a
highly energy efficient CMOS chip, and most of the other
circuits can be designed as such.  Without the power drain of
something like an Intel or 68K chip in there, the device could
run for a long time on batteries.  The Psion computer is neither
Mac nor PC compatible, but it is portable, lasts a long time
on batteries, and continues to sell well.  GS programs are far
superior to Psion ones.

-Scott G.