From: Dave Haynie 
Subject: A couple of questions about PPC and PIOS one...

And:  I hope things will finally stop screwing up...
      I have feeling that Apple's management is the primary cause of these
      PIOS problems.

Dave: Well, pretty much. Maybe it was a bad idea to get into the MacOS
      business, but really, with most companies, when they make you a promise,
      they keep it. Without that, it becomes practically impossible to conduct
      business. Apple clearly put a hurtin' on us, and I suspect our small
      size, ties with UMAX, and distance kept us from being completely wiped
      out. They clearly killed Power Computing, $100 million purchase or not.
      And our "technology partners", Motorola and IBM, did absolutely nothing
      to change the outcome, or even improve the market for PPC at all. Which
      of course explains why we're using an x86 clone (everyone bends over
      backwards to get you to use THEIR clone, knowing full well you can go to
      someone else any time) and Liunx (no one can take that away) in the
      Met@box.

And:  I hope that PIOS One will be resurected again when Amiga OS 5.0
      rolls out.

Dave: We do plan to pick up work on it again once the Met@box work is complete,
      hopefully this fall.

And:  Your prognosis that AI will be SW only development company seems to be
      very true...  Are you satisfied with the direction it has taken?

Dave: Well, in-so-far as they just do software, sure. The only way to make an
      OS in this day and age that can grow is to make it licensable by
      everyone, and then NOT compete with everyone licensing it. That's
      apparently what they're doing.

      As for the rest of it, who knows. If they're really fixed on requiring
      this magic MMC, I think they're doomed anyway. Fixing on specific
      hardware details was perhaps the only way to make the A1000, but it
      should have been killed in 2.0 if not before. It certainly held back
      every system from the A2500 onward. Today's wonderful MMC could easily
      be tommorrow's boat anchor. Now, I have no problem with them specifying
      a specific minimum level of performance, even based around some
      weird-ass MMC or another (hell, Microsoft is always specifying minimal
      levels of acceptance, their PC9x-specs for example), as long as I'm free
      to do better, with other hardware.

And:  Whats your opinion on the future of PowerPC family processors for
      desktop systems? (with powerful AltiVec technology and considering
      the opinion of Mick Tinker on Access Inovations web page)

Dave: AltiVec, if used well by PPC compilers (and it's not like this is rocket
      science, vectoring compilers have been used in supercomputers since the
      Cray 1 days), will make the PPC the fastest desktop chip around. On the
      other hand, there's little evidence Motorola and/or IBM are doing
      anything to make it an other-than-Apple thing. I mean, there was a time
      when 4-6 million chips a year was PC quantity, but today, you need 10x
      that to keep good margins going. Sun's doing a good 4-million SPARCs a
      year, and charging a hell of a lot more for them. If they don't have the
      desktop, they can't keep the best chips for the desktop going.

And:  And considering the fact that CHRP is probably dead platform (murdered
      by Apple!?) and probably the only CHRP computer will be PIOS one...

Dave: IBM already ships a CHRP system, I think it's called the F50. Anyway,
      that's invisible to personal computer users anyway, since it's shipped
      in their workstation line. But really, today, the One would go out with
      Linux and BeOS, basically into the workstation/hacker market anyway, and
      be nearly as invisible. Ok, perhaps we could hire some Swiss to do a
      hostile port of the MacOS or something (and we would, we're not weenies
      like Motorola), but that might not be enough to build the big numbers.

And:  ... considering that (as Mick has truly forecast) IBM is moving to
      IA-64 (they announced porting AIX to IA-64 if I'm not mistaken)

Dave: IBM isn't "moving" to IA-64. Think for a minute -- IBM is one of the
      largest x86 system suppliers, and that whole market will, one day, want
      IA-64 systems. Windows NT within IBM is at the same development priority
      as AIX, for tools (below Windows 95, above OS/2, though the OS/2 people
      downplay that). IBM is just one big honkin' company. Their workstation
      division may deliver IA-64 machines if there's a demand, but they're
      already saying that this year's Power3 machines are faster than Merced
      will be. They're very much in competition with IA-64 there, and probably
      ahead. I didn't hear the AIX rumor, but if they don't port, they'll be
      the only UNIX that isn't on IA-64. But how many aren't on x86? And yet,
      Solaris on x86 did nothing to boost Solaris, it's only become the most
      popular UNIX by selling SPARCs. There's also a rumor IBM will use Alphas,
      but who knows. Their divisions don't cooperate very closely, which is
      exactly why OS/2 lost out to the inferior Windows -- it was easy to
      think, "hey, if half of IBM and all of the rest of the industry won't
      support OS/2, why should we"?

And:  Sad that with all those perfect inventions as copper and SOI8 chip
      technology PPC CPUs will probably disappear from the desktop market.

Dave: IBM's applying it first in spins of the PPC750. Now, whether that's on
      your desktop or not, I can't say, but that's certainly where it's going.
      They aren't exactly making IA-64 chips, eh? Now, it's quite possible
      those 1GHz PPC750s will only find a home in phone switches or other
      embedded markets, I can't say for that. Again, if IBM and Motorola can't
      answer the desktop question except to say "Apple", is there a real
      chance? iMac may answer the Mac faithful with a machine they don't mind
      buying (I suspect, like the Amiga faithful, they as a group have more
      advice than cash to offer), but I don't count them back in the game
      until they have some real longterm strategy. And at least for me, that
      would require independence between OS and hardware.

And:  Your opinion on AMD and Motorola aliance?

Dave: Good for both, but not why you think it is. AMD needs better chip
      technology to keep up with Intel. Since they can't possibly match the
      large bags of cash Intel spends on each design, the process tweaks, etc.
      they'll make good use of Mot's copper interconnect process for much less
      money. Mot, on the other hand, is looking bad in their strongest market,
      the embedded CPU market, because they don't have flash ROM technology to
      mate with the 68xx, 683xx, ColdFire, MPC8xx, etc. chips that really
      should have on-chip flash. They get that in exchange.

      Whether it goes one micron beyond that, I couldn't say.

And:  Could it also mean some 'alternative' future for the PPC family or
      even creating some all over new CPU family -- maybe an alternative to
      IA-64 HP-Intel architecture?

Dave: Unlikely.

And:  (AFAIK IA-64 should execute x86 in HW somehow)

Dave: Opinions vary. It's interesting to note that Microsoft now has put DEC's
      FX!32 technology (the dynamic emulation/translation system that runs
      Win32 binaries for x86 on Alpha NT machines) in the NT 5.0 kernel. Given
      Intel's deep pockets, there's every possibility they have something like
      that in mind for IA-64. They have totally sidestepped the "how" of x86
      code execution at every step of the way.

And:  (And AMD has/had also some advanced but probably experimental only RISC
      superscalar CPU Am29000 line using techniques as renaming, reordering etc.
      -- I suppose they used the results of this development in K5 and K6 line)

Dave: Well, they used their 29K experience in the K5, I suppose, but the K5
      was a weak chip. The K6 was a result of their acquisition of NexGen.
      Interestingly, NexGen was the first to deliver a Pentium-II-like CPU,
      some four years before Intel did. Their Nx586 uses a backside L2 cache,
      it was the first CPU to dynamically translate x86 instructions to
      multiple, RISC-like instructions internally, etc. Also, not surprisingly,
      the first x86 clone to outperform an Intel chip at the same clock speed.
      Their nearly went out of business, though, simply because they didn't
      use the Pentium bus, so they had to provide the whole system design, not
      just the CPU. Too much for a small company, and it made it all the more
      difficult to convince folks to use the Nx586 (I have one in my lab).

      The K6, of course, is faster than the PII at the same clock speed in
      Integer, and though not widely known, also on single floating point
      instructions. The PII, though, has a fully pipelined FPU, the K6 doesn't
      (cost reduction), so the PII is faster on FPU-intensive stuff. Also,
      Intel's the only one doing SMP chips in the x86 world, unfortunately.

And:  Your general opinion on media processors? (ok, now talking only about
      those with no attemp to supply general purpose CPUs)

Dave: I think they can be useful. The proper approach is to design the system
      as a system -- MPEG, audio, etc. are functional blocks, and behave this
      way in software. If you have a low cost media chip for, say, set top
      machines, use it. Such a machine is probably going in the livingroom,
      and could well be decoding MPEG-2 streams as a DVD player, a DTV
      receiver, or for advanced gaming. It may not make sense on higher-end
      systems, especially if they're not likely to be used in this way much --
      would I better spend the money on a faster graphics card, and spend 20%
      (or whatever) of one CPU to decode MPEG if necessary.

And:  WHO do you think will design and architect the OS5.0 platform (HW
      references) if Amiga, Inc. will be SW only company?

Dave: That's a good question. I really don't know.

And:  Did you plan this to happen to IC (AFAIK it was also your idea, maybe
      even only your idea?)

Dave: They still call it ICOA because of my "Industry Council/Open Amiga"
      paper, which did get the whole ball rolling. While I had the idea back
      in '94, it was actually folks like Giorgio Gomelsky and a few other JMS
      people who pushed my buttons and got me to summerize things again. I
      know Carl Sassenrath, independently, had tried to get something similar
      going once.

      My intention for the Industry Council was pretty much this kind of thing
      -- the council needs to address Amiga Industry Standards, working with
      the Amiga owners (eg, Amiga, Inc) when they cover a thing, working
      independently on questions they don't care to address. This could be
      something relatively minor, or something as major, I suppose, as a
      reference platform. The other thing, of course, is coordination within
      the industry. In my position on the edge of Amiga hardware at Commodore,
      I saw most of what everyone else was doing, and the fact that it usually
      the case that everyone was inventing the same wheel at the same time.
      You wouldn't expect Phase V, VillageTronics, and the four or five other
      companies all working on Zorro to PCI bridge solutions to cooperate --
      after all, they're in competition. But really, all that actually happens
      here is that five lesser designs are created, something useful for
      everyone in the industry. A single VHDL design, synthesizable by anyone,
      would have freed these folks up to do other, more interesting stuff.
      Which, in turn, benefits the Amiga industry as a whole. Commodore, as
      well as most of the third party companies, refused to take the big
      picture view as The Amiga Industry being more important than any one
      component. That, in fact, is really what you get when you split software
      development off as an independent company (examples: Microsoft, Linux;
      nuff said) -- no one component becomes important enough to kill the core
      hardware business, and thus, the software always has value to someone.
      The problem in restarting the AmigaOS would have been fairly minor,
      comparatively. Needing, today, the whole custom chip infrastructure,
      based on outdated technology, has been a killer.

And:  As I know Amiga, Inc. has contacted also PIOS Ag. because of developing
      and producing the AmigaOS 5.0 platform computers...

Dave: Nothing seriously. I have had no direct contact with Amiga, Inc. And I
      am, after all, VP of Technology. Any serious discussions, I would have
      been involved. But I do have people sending everything I say on some of
      the Amiga mailing lists to the right people, supposedly, at Amiga, Inc.
      So at least maybe they'll get the right idea, but who knows.

And:  So the logical question is what will happen to PIOS one project: If OS5.0
      will be HAL based (eh, *should be*), will PIOS write your own HAL for PIOS
      One?

Dave: CHRP defines about 90% of what I would call a system HAL. You could
      write an OS to talk to that directly and need only a very small
      collection of hardware-specific routines, or supply an OS-level HAL,
      like Windows NT does. I don't much care where it comes in. If Amiga,
      Inc. does a PPC AmigaOS with HAL, and publishes the HAL, we'll write one
      and make sure the AmigaOS runs on the One.

And:  Anyway, what is the future of PIOS one project? (counting with the not too
      much certain future of PPC CPUs for desktops, etc.)

Dave: We do plan to move ahead with it, but it's been tabled until the Met@box
      is completed. There's much more support for the Met@box right now in the
      company (well, they did change the name to "Met@box Infonet, AG" --
      though the name change was necessary, due to a trademark screwup made by
      our lawyer in Germany).

And:  With the flexible CPU-module architecture, will you move to some other
      CPUs? (Merced?)

Dave: No plans to yet, but it's absolutely a possibility. Merced? Not an issue
      for this millennium....

And:  Or will you adapt the PIOS one concept (and CPU module) to work with
      THAT MMC?

Dave: That's a question, especially since they haven't made it at all clear.
      Logically, an MMC belongs on the PCI bus, not in the CPU card. Of course,
      logically, the OS runs on a CPU, not an MMC. So we'll have to see which
      part of the puzzle they're not telling us, before making any kind of
      plan here.

And:  Do you plan 100Mhz local bus for PIOS one?  Does exist such a Grackle
      chip which supports this speed?

Dave: The new (new for the changed CPU module connector) single PPC module was
      done using Motorola's 100MHz design rules. Grackle at Rev 4 can run to
      100MHz, though the timing is very, very tight. We haven't made this
      module yet, since the four-CPU module (done by Thomas) overlapped my
      work on the new motherboard, we're starting up with the x4 module, and
      adding the x1 module once things are fully debugged. I would also like a
      PPC750 and/or a Max module, but those will come later. You aren't
      getting 100MHz with four CPUs.

--
Dave Haynie  | V.P. Technology, PIOS Computer |  http://www.metabox.de
Be Dev #2024 | NB851 Powered! | Amiga 2000, 3000, 4000, PIOS One