NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/comp/unix/pcclone/32bit/1311

Internet Message Format | 1993-01-24 | 6.4 KB

Xref: sparky comp.unix.pc-clone.32bit:1311 biz.sco.general:5541 comp.unix.sys5.r3:394
Newsgroups: comp.unix.pc-clone.32bit,biz.sco.general,comp.unix.sys5.r3
Path: sparky!uunet!zaphod.mps.ohio-state.edu!news.acns.nwu.edu!nucsrl!ddsw1!karl
From: [e-mail] (Karl Denninger)
Subject: Re: PC Unix/Xenix vendors
Message-ID: <[e-mail]>
Date: Mon, 25 Jan 1993 02:25:10 GMT
References: <[e-mail]> <[e-mail]>
Organization: MCSNet, Chicago, IL
Lines: 110

In article <[e-mail]> [e-mail] (Geoffrey Welsh) writes:
>[e-mail] (Karl Denninger) writes:
>
>> Correct.  This is the reason I suspect the INTEL architecture is in trouble
>> over the long run.  Certainly, they have the installed base now.  But these
>> "installed" systems do not play well together, they're unstable (ask any DOS
>> user) and to get away from that is just too damn expensive.
>
>   Actually, I look at it the other way:  CISC CPUs are slowly narrowing the
>performance gap with RISC through parallel execution, smart pipelines, etc.;
>while other 'power CPU' manufacturers slug it out on workstation volume, Intel
>will be able to manufacture one chip that squeezes into the workstation market
>and keep their cost down by convincing the DOS people who fancy themselves
>'power users' to buy the same CPU for their PC.

Perhaps.  However, the Pentium is not yet out, and yet there are other
RISC-style CPUs which <are> shipping that have the parallel execution
threads internal to their architecture.  That is, Pentium may be trumped
before it is released.  It is almost certain that it will break no new
ground in performance -- by any other measure than PCs.  HP 'Snake' machines
already get ~70 MIPS performance; the best I've heard for Pentium is near
100 MIPS of integer performance.

It is entirely possible that someone else will ship a 100 MIPS CPU before
Intel.  Besides, you can easily get that today with the Sparc parallel
machines, some of which you can get on the desktop now.

>> The real problem so far has been that to get a "real" computer was a $10k
>> proposition.  Now, with the Sparc Classic, it isn't any longer.  Consider
>> that the clone makers will go gunning for Sun's bacon here, and I bet we
>> see $2500 Sparc compatibles within 6 months.
>>
>> Note that the compatibles have <always> come with an operating system too.
>>
>> Now just HOW is the Intel machine going to compete with <this> in that
>> marketplace?
>
>   Simple: within six months we'll see Pentium machines also starting in the
>$2500 ballpark.  They'll sell because of DOS and its plentiful cheap
>applications, and Intel will recoup their development costs a whole lot faster
>than the SPARC group.  Those of us who appreciate the finer architectures
>(even if we can't always afford them for our home computers <grin>) will
>lament DOS yet again.

Not a chance.  Pentium machines won't hit the $2500 ballpark for quite a
while.  You're forgetting a number of things which are going to go into that
equation:

1)    The Pentium, if it follows previous Intel philosophy, will
    originally be priced at a few <thousand> dollars a chip.  Further,
    you need the support logic, which isn't going to be there
    immediately, and will also be expensive.  The clock speeds these
    things will run at will significantly tighten the engineering
    requirements; only the best designs of today will qualify for a
    working system.  Therefore, you're going to see Pentium first in the
    $15,000 network servers.  My prediction is that Pentium machines in
    "reasonable" (16MB RAM, etc) configurations won't break $5k for at
    least 18 months.  And that is <without> a frame buffer or disk.

2)    Pentium will be a no-second-source chip.  This means that Intel
    is going to bet the farm on it as the "future", therefore, they will
    keep prices high.  They didn't make the same mistakes this time
    around with licensing; there won't be any competitors for at least a
    year in the processor area, and I would bet Intel can <successfully>
    sue their imitators this time.

3)    <Someone> is going to come up with an effective software/hardware
    emulation of the DOS "working set" within the next 12-18 months on
    non-Intel processors.  As the hardware gets better, and faster, this
    becomes more viable.  "SoftPC" on the MAC <works>, its just slow.
    But what if it had 10x the processor power to work with?  This will
    come to pass.  This is a shot-in-the-dark prediction, but the
    company that manages this is going to blow Intel's chops off --
    therefore, the return on risk is incredible.  I would bet this shows
    up on one of the Sparc systems first, and probably within the next
    12 months.  With effective 386/486 <enhanced mode (paging)> support
    available at, say, 486/33 speed, there will be less reason to run
    DOS.  Note that the DOS emulators are <already> available on the
    Sparc machines.  The problem has always been acceptable performance.
    This problem is rapidly disappearing as technology improves.

>> Pentium isn't an answer if it is a monopoly chip.  If Intel insists on
>> having all the marbles they may find themselves with none of them.  Pentium
>> will be fast, yes.  But a multiprocessor Sparc-based system will be
>> available by that release date <on your desktop> which will make Pentium
>> look like small potatoes, and will likely be cheaper as well.  Certainly the
>> Sparc people can compete price-wise.
>>
>> That's my prediction for 1993 (if Intel ships in '93 ;-).
>
>   I disagree with your prediction, but I hope that you're right.  Frankly, I
>want to see an end not only to DOS and its real-mode limitations and its
>multitude of incompatible and/or unstable workarounds, but also to the PC
>architecture, which imposes unnecessary problems on operating systems that
>might otherwise take better advantage of the CPU that's hiding on that PClone
>motherboard.

So do I.  Time will be the judge, as it usually is ;-)

Frankly, my hope is that one of the Sparc people (TI perhaps) incorporates
some "special" instructions intended to make PC emulation easier.  With
that, and some of the new superscaler Sparc designs, you could have
"enhanced mode Windows 3 / NT" compatibility at the <binary> level running
under NeWS or X11.  Once that happens you automatically get the command line
(straight DOS) stuff essentially for free.  Then INTEL has no monopoly, as
the choice of software no longer dictates a machine architecture.

--
Karl Denninger ([e-mail], <well-connected>!ddsw1!karl)
Data Line: [[phone]]