🚀 go-pugleaf

Mail from Dave Haynie:

> Multiprocessing is THE way for AT to distinguish themselves from all
> the other CHRP-machines, and gain a competitive advantage. Go for it
> AT!

I don't think so. Because it's awfully likely much of the forthcoming
CHRP hardware will support multiprocessing. The Motorola glue chips
support at least two processors (the new CHRP version handles at least
two 603s or 604s and L2 cache). I don't know SIS or any of the other
PC-based glue chip companies supporting PPC will do, but if Motorola
has managed to set the standard, it'll be impractical to build a PPC
system with off-the-shelf parts that at least couldn't support
multiprocessing (eg, the designer isn't forced to make room for that
second CPU). Apple is probably rolling their own system chips, and I
think Amiga Technologies must for the low-end too. Beyond that,
though, I suspect everyone else will use standard parts. 

Now, multiprocessing support in the OS isn't a given, but it's coming
fast. Be, clearly, is doing it nicely. Windows NT, as well, supports
multiprocessing, though it's apparently clumsy next to the BeOS. IBM
has a multiprocessing version of OS/2 that's been shipping for the PC
for awhile, so there's reason to expect a multiprocessing OS/2 for the
PowerPC. Same goes with Solaris (Sun's UNIX); it's been
multiprocessing for years, it would be reasonable to expect that in
the CHRP version too. Apple isn't currently multiprocessing, but
there's already talk of a 3rd party version of their "Thread Manager"
that supports a somewhat limited multiprocessing on today's PowerMacs
(or perhaps PowerMac clones, since Apple doesn't have a dual-processor
machine of their own). Copland is supposedly going to incorporate and
perhaps even improve this idea. 

So, basically, multiprocessing seems to be standard on the PowerPC
platform, if not necessarily in every system or OS. 

> Single processor machines are a perfect way for CPU-makers to earn more and 
> more money, nothing else.

It's not like they're stealing or anything. New CPUs cost a fortune to
design. Intel spent something like 100+ million developing the
P6. Sure, Kevin Costner can spend more remaking "The Road Warrior" at
sea, but for mortal businesses, that's big bucks. And representative
of how much each new chip has to cost to reclaim that expenditure and
stock up for the next generation, which will cost even more. 

New chips are expensive because of this, and because there's always a
pent-up demand and a limited supply when they're introduced. But CPU
companies don't mind multiprocessor systems at all. The only way
SGI/MIPS can claim to have 3-4 million installed RISC processors
(which they say is the largest installed base, though most RISC
processor companies tend to warp facts their way too) is that they
sell SGI boxes with lots of CPUs in them. Motorola's learned from
this, and from the Intel thing (supporting the whole motherboard), so
they're looking to sell you a couple of CPUs, some cache RAM, and the
system glue chip for the PowerPC. For the 680x0 generation, they sold
you just a CPU. High-end CPUs don't go away just because of
multiprocessing, but adding good multiprocessing support to your
system chips takes some of the heat off the design cycle on the next
part. 

> I don't know yet which will be the next computer I'll buy, but it definitely 

> won't be a single processor machine. If the PowerAmiga will do 
multiprocessing 
> and will be out in time it's a good candidate. Otherwise I would bet my 
money 
> on the BeBox. Best would be to get both - a BeBox beginning '96 and a 
> PowerAmiga in '97 - hmm.. now i just have to look for the money... someone 
can 
> lend me a few thousand $$$$ ;-) ? 

If you want to hack, a BeBox would be fun now. Otherwise, hold on to
what you have for most of '96. You won't see many great Be
applications until early 1997, by when the PowerAmiga should at least
be on the show circuit, and while the CHRP market won't be quite
settled down yet, at least most of the cards will be on the table, and
you'll have your choice of CHRP systems and at least some shipping
OSs. This is the advice I've been handing out. 

> I didn't mean that there should be no more AmigaOS, but that AT should take 
> the BeBox-hardware, and all the BeOS functions which are supporting the 
> hardware, and build the PPC-AmigaOS with these ingredients, therefor coming 
> out with a PowerAmiga with AmigaOS much earlier than if they start to cook 
> their own soup.

Nope, that's not it either. The low-level hardware functions are
specified and managed by the CHRP hardware abstraction layer
(HAL). That's not done in the BeBox today, but it will have to be for
real CHRP machines, and so will the AmigaOS. That gets your memory
configured, your hard disk booted, your display initialized,
etc. Going much beyond that level, you're either on your own, OS-wise,
or you're hosting as virtual OS under someone else's OS. The former
might take a little longer to write (though realize, much of the
AmigaOS can run in emulation for much of the development process, with
modules being independently brought native), but the latter will be
less efficient. While both OSs have the same system services, it's
very likely that making a Be task look like an Amiga task, etc. to a
program is going to be very inefficient. Especially because much of
the AmigaOS API (and much of the reason for its speed) is based on
"hot" structure access, not function or object APIs. The Amiga OS
expects a Window to be a Window, and a offset X there's a RastPort,
and off that is a message port, etc. This could very well be a
nightmare to get hosted at all under another OS; I think you'd wind up
doing the whole AmigaOS port anyway, and just slipping one extra
software layer in-between to slow things down.

> That's true, but I wasn't thinking of the $500 market segment for the 
> BeOS/AmigaOS solution. In the near future I don't see any chance for a 
> PowerAmiga in this price segment, 

I do, and more to the point, I think that's a prime target for
AT. Sure, an A4000-class system is easier. I could build one with
off-the-shelf parts in 4-6 months (just me, not a whole design team),
though you'd probably have to give up 800K floppies, which would a
problem (though there are some avenues yet to be explored). A cheap
system would require high custom integration, that simply takes
longer. 

> and I believe that the customers for a $500 machine won't care too
> much about the CPU being a 68k or a PPC (as well as if it's
> CHRP-compliant or not).

What they will care about is performance. And games, good ones that
don't leave them wanting too much for a Playstation or Ultra64. A
cheap internet hookup means, so they aren't suckered in by whatever
Sun or Oracle come up with in sub $500 netboxes. The means to that end
isn't that important, true. But the CPU is just a small piece of this,
and I supect a PPC system may be easier to build this way (especially
given that the 3DO M2 system is based on the PPC). Also, if you're
going to bite the bullet and do a proper $500 computer (realize the
A1200 is more like $750), why not use the new technology. It might be
harder to start with, but you don't have to do it all over again a
year later.

> I was rather looking at the >$2000 market segment and I don't expect too 
many 
> new customers to invest this amount into a 68k-machine.

Certainly not. I believe AT can make a Power Amiga in the A4000 class
for much less than today's A4000. Yes, it will have to use lots of
standard parts.

> IMO it would be fair enough by AT to offer beginning/mid 96 a PowerAmiga 
with 
> 68k-AmigaOS emulation and tell the customers: yes, this won't run as fast as 

> it could run natively, but on the other hand it's still faster than on an 
> 68k-machine, and you don't have to change the hardware in one year. 

Sure thing, except that there won't be a PowerAmigaOS in beginning/mid
1996. Would be nice, won't happen. The hardware really isn't the
issue; like I said, they could use a Motorola (or, yes, Be)
motherboard, the problem is the OS. Don't forget, Apple took three
years for their MacOS port, and they didn't have to rebuild the whole
company while doing it. 

>> On the other hand, they have to pick some PPC system to start
>> porting to, why not a BeBox?

> Yah, why not? Can AT hear this? ;)

I'm sure they have thought of it. I have reason to believe they've
been getting some good advice lately. 

					-Dave Haynie