--------- START OF FORWARDED MESSAGE ---------

From System ....: Internet
From Conference : comp.sys.amiga.misc
MessageID ......: <335ec578.1324198697@news.jersey.net>
Originally from : Dave Haynie <dhaynie@jersey.net> <609399483>
Subject ........: Re: Results From the Amiga Research Project

On 23 Apr 97 20:02:42, vlangber@online.no (Vidar) wrote:

>On Tue, 22 Apr 1997 22:21:23 Dave Haynie wrote about "Re: Results From the 
>Amiga Research Pr":

>> But they have hundreds of things
>> more important to do right now than starting a multi-million dollar,
>> multi-year development project. 

>Finishing the Hombre wouldn't be a multi-year development project.

No? Well, maybe you know more about it than I do. I really don't know.
I do know that no chips exist, that no final design was taped out,
that Ed made good progress on the design, but it wasn't complete.
Maybe he finished it. 

There are two good reasons you could finish Hombre sooner than AAA.
The first one is simple: Hombre's a new design, the only compatibility
target is the spec. If it's working 99% to spec, you can probably
change the spec. Can't do that where compatibility is an issue. The
second one is that you probably could hire Dr. Ed to complete it, as
long as travel wasn't important (he doesn't fly). There's little
chance you'd track down and get back the rest of the AAA team, even if
it was a good idea.

>>. But after
>> that, I looked into the graphics chips that were available, and three
>> years ago, you could get chips for $25 that did faster graphics than
>> AAA: higher/deeper resolutions, higher scan rates, faster blits, etc.

>But would they beat the Hombre? I really don't think so.

Opinions vary. Some of our graphics guys felt that the Sony
Playstation's graphics engine had it over Hombre in some/many areas. I
don't have enough real benchmarks on Hombre to judge this. Some of it
will be application dependent, certainly. At least you can ask the
question without, on the surface, sounding silly. Also, based on the
high-level design, it's possible things could be updated faster, once
the target process were known. 

For the record, I do advocate looking into Hombre. 

>> you can do games with a simpler API). 

>Do you mean one like Apple's RAVE? Or Direct 3D?

I was thinking of Be's 3D Kit. The basic idea is that OpenGL has to
render things true, or it's wrong. A game kit can cut corners, like
most of the 3D games you see today do, especially on consoles.

>> Phase V's design is on steroids, perhaps, and they're using the
>> Commodore model. But they're not building Amiga chips. Neither was
>> Commodore in the end, Hombre was the last project going, and it was
>> not Amiga compatible in the slightest (by design, of course). 
>
>When you say Amiga chips, are you refering to chips that are register compatible 
>with AGA etc.? To me it seems that all chips designed for the amiga should be called 
>Amiga chips.

I don't think so. I would claim that are used in Real Amigas are, by
default, Amiga chips. Anything based on a similar architecture might
beg the qualification. At Commodore, Hombre was shouted from the
rafters as NOT being an Amiga chip. It's an Hombre chip. I did plan to
use it, optionally, in future high-end Amigas. But at that point, any
graphics chip would have worked. Real RTG was necessary for anything
from AAA onward. 

Quite frankly, one could argue, and they'd be right, as well, that
it's not an Amiga chip if it wasn't designed by Jay Miner, or at least
decended directly from his architecture (like the Lisa chip, for
example, clearly a better Denise, but designed by Bob Raible). 

This goes back to my point about Gateway. They can legally call
something "Amiga", even if won't pass anyone litmus test of
"Amiganess". What we've all grown to know and love as the Amiga isn't
a magic formula anyone can whip up and renew, it was the serious,
clever, and fun as hell efforts of a number of very talented
individuals over the course of ten years of evolution. MOST of that
evolution took place in software, and that's the "genetic material"
most easily transferred to some other machine. 

And as well, that's the part that makes the beast. You can't tell by
looking if I have one or two hearts, or two, four, or six chambered
hearts, or whether my heart is in good shape, or plaqued with fat.
You figure I'm a human by the surface stuff. It's the same with the
Amiga: the hardware is the organs; you assume something about these
organs when you see the Amiga nameplate. Usually you're right, but
there are these really fast machines out there, which are pretty much
the same as always, only way faster. Those bad boys have alien six
chambered hearts, courtesy of Phase 5 or VillageTronic. They still
have the same skeleton, same flesh, same musclature, same mind. Then
again, a few have been training their brains on one of those Altair-IV
brain test machines (got '060 card), or even alien brain transplants
(PowerUps). 
  
>By the way, everyone here sems to be to fixed on specific performance in certain 
>areas between off-the-shelf versus custom chips. I'm sure your are right when you 
>state that any custom chipset would be surpassed by off-the-shelf chips in specific  
>areas within 6-12 months. But to beat a nicely done custom chipset in most areas 
>would require several chips. 

No reason it should, unless you want it that way. The competition is
using large scale integration at least as good as anything Gateway
could contract, very likely something better. You're up against CHIP
COMPANIES here, not other systems houses like it was in '85. It's
exactly like doing your own CPU.  All-in-one chips are available today
with graphics, sound, MIDI, acceleration, 3D, MPEG, modem, etc. The
only time you really want something that integrated, though, you're
not building modular systems. Even then, it's probably a bad idea.
Read on...

>The custom chipset (and the OS) initially made the amiga popular in the video and 
>graphics industry. 

It was an entirely different world 12 years ago. The Amiga taught lots
of people some good lessons. 
     
>Let's say that Gateway decides to finish the Hombre. It would probably be released 
>in 12-18 months. Hombre's 3D stuff might be surpassed by then, but nothing will 
>beat an Amiga based around this chip in price/performance.

Or would it? Why could Gateway necessarily make Hombre cheaper than
someone else could sell a chip to Gateway with similar features, at
least reasonably so. Gateway has to get volume on this part; it's not
a trivial engineering effort, and you must pay for that engineering on
each chip, over the lifetime of the chip. If they sell 100,000 Amigas
(quite a lot to expect for the first year or more), and spend $10
million to develop the chips, that's $100 per chip exclusive of the
chip's actual cost. That cost won't be great, either, in those
volumes. The competing chip, shipping in 10 million unit volume, let's
say, is only $1 per part in R&D. Probably less, since it's based on
the architecture from last year's 10-million-seller, so it cost less
to develop in the first place. See the trouble here.

So let's say Gateway really wants this to succeed. Maybe they can put
this chip in every PC they make, so now you have 5.1 million chips per
year, much better, that's less than $2 per chip. But the new chip has
to be good for Windows machines, at least as good, in Windows, as what
the other guys are doing. Does this make the best Amiga chip?
Sometimes. If there's a conflict of interest, which lobby wins out?
Well, at a 50:1 volume margin, that looks like the PC to me. If the
Windows drivers work, will the ship the chips even if there are Amiga
features not working? You bet your bippy they will. See the trouble
here.
  
>The Hombre as I see it got almost none of the traditional drawbacks of a custom 
>chip:
>
>- it's cheap. The target price was $45, using 0.5 micron technology, but with 0.35 or 
>0.25 technology, the price should be significantly lower.

You can't get 0.25 or 0.35 technology under contract. The only way is
to get a chip company to be your partner. That's how Hombre was
supposed to work at C=, when 0.5um technology was the leading edge: HP
would use the chip for PCI-based machines they make, and make the chip
for both of us. This is how start-up Chromatic got their MPACT chip
made (Toshiba and LG Semicon make and sell it). Maybe it could be
done. These days, $45 can buy you lots of graphics chip.
Better-than-AAA performance is around $15.
         
>       - Because of that it should be pretty easy to implement more features like JAVA 
>acceleration, Wavelet video incoding/decoding etc. 

This is the kind of thing that programmable chips run in software. The
real question is whether the Hombre design is right for this. The
PA-RISC used in there was decent as a GPU, maybe, you'd have to
evaluate its performance on various things like MPEG-2 decode, etc.
Most of the on-chip-processor graphics chips you can get today use
some version of a VLIW architecture, so you have maybe ten processing
units running at once. We're talking 2-20 BOPS here, not maybe 50MIPS
or whatever that PA-RISC does.
 
>       - I believe that the cutom HP PA-RISC in the Hombre is based on the PA-7150. 
>Wouldn't it be pretty easy to implement the custom stuff in a faster processor like the 
>PA-7300LC (or even PA-8000 for the high end)?

I think the design was based on a more generic netlist from HP, with
many improvements done by Dr. Ed. This was, however, slower at
conventional work than a PA-50, I believe, The idea was, the two
Hombre chips, plus whatever I/O you needed on the PCI bus, gave you a
CD64-type game machine. Add a PA-50, or a cheap PPC, and you'd have
the A1200 replacement. Put it on a card, and it would plug into the
PCI bus on a midrange or high-end machine. 

You can't get a better PA-RISC, because HP doesn't license any of
their high-end designs or sell these chips to third parties. ,That's
why the PA RISC was unacceptable as a desktop CPU. And I suspect the
PA-8000 runs about $1000 a chip, if you could buy one. Not exactly
what we're talking about here. 
      
>       - The Hombre display chip is small, so it should be pretty easy to make a new 
>one with improved performance. The memory subsystem could be upgraded to the 
>new memory you wrote about the other day (synclink or something).  

It used FPM VRAM, today you might use EDO VRAM, WRAM, or Synchrous
DRAM. Maybe SyncLink next year. Rambus is nice for some things, but Ed
felt the Hombre would be jumping around in memory too much for this (a
problem you can solve with enough on-chip buffering, but that adds
complexity and cost). 
      
>Finally, look at the consoles. Without custom chip designs there would not be 
>anyone near the power they have today at that price.

They sell millions of these a year, at least when they peak. But even
there, they're not entirely custom. Most use off-the-shelf processors.
The Sega Saturn's audio system uses off-the-shelf 68K, Yamaha DSP, and
512K of memory. If the machine lives long enough, they probably go
back and integrate more. Realize also, though, that you're paying
close to cost for these things, compared to PCs with real margins 

> Nothing can beat a highly 
>integrated machine/console on price if done right.

Of course not -- you can't make money selling a computer at cost, you
can a game console (in fact, since everyone else does, you MUST if
it's a game console). 


Dave Haynie      V.P. Hardware Engineering    PIOS Computer
haynie@pios.de     "...no RISC, no fun"

---------- END OF FORWARDED MESSAGE ----------