Media Chip (fwd) Re: INTEL? NE! Tip na super-chip: CTETE....

[Bob Stifter <stifter@mira.cz>]

>Ctete: http://www.chips.ibm.com/microdesign/vol2_no1/  ---> clanek Q&A
>with Henry Massalin

[strih strih strih]

> > Amiga, Inc. also announced plans to unveil a personal computer that is an
> > industry standard architecture solution.  The new system will provide a
> > bridge to the next generation of Amiga technology.  

> I can't see the "68k and PPC" processor decision mentioned here. 

Clearly, "68K" is implied on any new system under the "emulation"
category. 

I had pretty much guessed they would go to x86 this year. First of all,
they don't have the prerequisite "hardware visionary" at Amiga, Inc.
Second, they're a division of Gateway, a company making truckloads of
money selling Pentium-based systems. Third, the real question is AmigaOS,
not Amiga -- they really should spend most of their effort on the OS.
Part of their ability to gain support for the OS is exactly what Apple
refused to do: the OS must extend beyond any systems they deliver. The
largest target out there is the x86 PC, like it or not. 

I have ten separate "rumor mill" emails claiming they have some interest
in a CPU no one's ever heard of, which won't be available this year. This
could be true -- new companies are always making new things. And while
that could be good, the danger is guessing about anything that's coming
in two years. They don't quite come out and say "new CPU architecture"
here, but they are talking about a new architecture some kind. If that's
some kind of new system architecture with x86 running it, you could have
that in November, there's no need for a two-year wait. If it's new
graphics chips, fine, but these ought to just plug into existing
machines. 

So what about CPU? Everyone thinks their faster-than-ever chips are
falling behind, these days, but they can't quite explain why. You can
look at VLIW DSP chips delivering 1.6GOPS (billion operations per second),
such as the TI C60 family. Or media chips deliving maybe 20GOPS. But
these are very special purpose. VLIW, in general, rarely if ever runs
close to it's theoretical maximu, and it's an architectural throw-away --
you change the instruction set with any major upgrade of the architecture.
Intel's "EPIC" technology is supposedly a way around this, but you're not
looking at these going for$100 or less in 1999. Media chips (like
Chromatic's MPACT, Philips' TriMedia, etc) are not general purpose, they
hit their peak only doing one particular type of operation. They coexist
with a CPU if you like (we just had that discussion), they never replace
it.

Cheesy old x86 chips still double in performance every 12-18 months, and
they're not overpriced so much anymore. The new "3D" instructions should
accelerate some floating point operations by a factor of four this year
-- making them faster than today's Alpha chips, at least on multimedia
data types, if not scientific computing. Of course, you have to recode
for this (and dodge various kludgeries, such as floating point register
overlaps) to get it. New code -- but only small bits, not the whole
system. If you think about AMD's 3D set on a 300MHz processor, you're
already talking 1.2GOPS, peak, for 32-bit floating point. That's not so
far behind that DSP, after all.

The Alpha itself is in flux, but it does look like AMD will be acquiring
all of the technology rights (they had previously licensed the 350MHz
21264 bus interconnect only). This might finally get lower cost Alphas,
but I still write Alpha off for the low-end, and they do have one thing
right in this announcement: an Amiga or Amiga-replacement system should
not be an expensive machine if you expect to build on the present
Amiga's installed base. You might replace some $500 machines with $1000
machines, but it's rare to replace $500 machines with $3000 machines.

I think PowerPC is the best technology choice (I can't answer for any
deals they may have made -- PPC itself began life not as the best
technological choice for Apple, but the best deal for Apple). First, you
have IBM pushing the technology ahead in clock speed, soon to rival
anything from Alpha. Next, you have Motorola's AltiVec, which some will
characterize as "MMX for PowerPC", but in fact, it's significantly
better. Based on a totally separate execution pipeline with 32 new
128-bit-wide registers, AltiVec is for general purpose number crunching,
it's not a type-specific kludge like MMX. For one, it supports floating
point, so a simple MAC instruction on a 300MHz processor is running
2.4GFLOPS. But also character vectors -- combined with sorting ops, an
AltiVec chip can do some kinds of string functions maybe 80x faster than
a normal RISC (including PowerPC) could via plain old byte operations.

Dave Haynie  | V.P. Technology, PIOS Computer |  http://www.pios.de
Be Dev #2024 | DMX2000 Powered! | Amiga 2000, 3000, 4000, PIOS One
      Buy my house! Take the tour at http://www.jersey.net/~dhaynie



StB



  * Bob Stifter --===***===-- TEAM AMIGA *
  * tel. +420 602 834287 -==- A1200T 603 *
  *         Zavolej mi do klubu          *
  * http://www.mira.cz/~stifter/HemP.htm *
  * HemPMagazine  By Skaters For Skaters *