EnigmA Amiga Run 1995 November/EnigmA AMIGA RUN 02 (1995)(G.R. Edizioni)(IT)[!][issue 1995-11][Skylink CD].iso/earcd/docs/lwml9504.lha/LIGHTWAVE-APR95/000433_owner-lightwave-l _Tue Apr 18 17:11:11 1995.msg

Internet Message Format | 1995-05-06 | 10KB

Return-Path: <owner-lightwave-l>
Received: by netcom.netcom.com (8.6.12/Netcom)
    id NAA18535; Tue, 18 Apr 1995 13:47:28 -0700
Received: from lois.dti.com by mail4.netcom.com (8.6.12/Netcom)
    id KAA01486; Tue, 18 Apr 1995 10:19:41 -0700
Received: from menage.dti.com ([192.84.116.10]) by lois.dti.com (8.6.9/8.6.9) with SMTP id LAA01412 for <[e-mail]>; Tue, 18 Apr 1995 11:39:59 -0500
Message-Id: <[e-mail]>
From: [e-mail]
Organization:  Deskstation Technology
To: [e-mail]
Date:          Tue, 18 Apr 1995 12:19:25 +0000
Subject:       Re: RAPTOR 3's Processer-Independent technology?
Reply-to: [e-mail]
X-Confirm-Reading-To: [e-mail]
X-pmrqc:       1
Priority: normal
X-mailer: Pegasus Mail/Windows (v1.22)
Sender: [e-mail]
Precedence: bulk

>Return-Path: [e-mail]
>Cc: [e-mail], [e-mail],
>        [e-mail], [e-mail]
>Sender: [e-mail]

Since I'm the architect of the Raptor 3, I though I might contribute
a few facts and a few of my opinions to this discussion, for what
they're worth :-)

>>My question is:  Does having the processor independent of the motherboard
>>really make the Raptor 3 immune to obselence?

Nope.  The goal of the Raptor 3 is not to make a CPU immune to
obsolescence.  I'm not smart enough to do that.  The Raptor 3 is
meant, however, to lengthen the usable life span of a system, at the
cost of replacing the CPU/cache module from time to time.

>CPU's are one part of a many variable equation. Performance of the processor
>is important, no doubt, but I/O bandwidth, memory design and many other
>variables play on the total throughput and life of a computer system.

This is pretty much true.  However, how much these factors affect you
is a function of the use to which you plan to put the system.   For
example, the performance of an on-line transaction processing system
is MUCH more sensitive to I/O bandwidth than a system used for
rendering, for example (technical details to follow).

> Two
>years ago the hot item was a 486/33 with a Pentium Overdrive slot. These
>systems supported the latest bus, VESA, and were upgradeable to future
>Pentium processors. Was your investment protected? Would you like that
>system today? Could you have predicted the availability of PCI, 72-pin
>SIMMs, EDO memory, new cache types, etc.?
>

Sure you could have predicted the availability of these things two
years ago.  Deskstation has been using 72 pin SIMMS since I came to
work here over 4 years ago.  If I remember correctly, I also had a PCI
specification in my hot little hands two years ago (although it may
have been closer to 18 months ago).  The interesting things on the
horizon right now are 64 bit PCI (seems to be fading fast) and double
speed 32 bit PCI.  If you buy a Raptor 3 today, you won't get those
things.

Having observed the high-end-PC/low-to-midrange-workstation
market for about 8 years, I conclude that a technology will begin to
ship in some interesting volume about a year after I see running
hardware at fall Comdex.  I didn't see any 64 bit or double speed 32
bit PCI products last year (although I was pretty busy and I'm
not omniscient) and I don't know for certain about the ship dates of
any Intel-based chipsets that support either of these protocols.
Since the Intel-based chipsets drive the market, I conclude that I
have at least a nine month window where PCI32 is the most interesting
technology on which to base a product and that PCI64 and PCI32/66MHz
won't have a significant base of add-in cards for at least another 18
months.


>Another variable is vendor stability. If you purchase the Raptor 3, you have
>a proprietary slot that supports future processor cards. Hence you are
>"locked" into DeskStation for future products to protect your investment.

No question about it.  However, you always have the option of buying
a system from our competition if we are not competetive on
performance or price.  Your investment in hard drives, add-in cards,
and memory SIMMS is preserved to the same (lesser :-)) extent that
our competitors preserve it.

>You must rely on their ability to develop hardware and software for diverse
>processor architectures. Not only must the hardware work, but a
>processor-specific BIOS and Windows NT Hardware Abstraction Layer must be
>developed and tested to make a functional system. How do they test their
>systems? Can they guarantee that it is bulletproof? How many engineers do
>they have working on it? Are these engineers versed in Alpha, MIPs, PowerPC
>and Intel Assembler?

The answers to that are (in order)

Our systems are tested with a 48 hour stress test that exercises the
system and exorcises any bad components.  As far as I know, this is
the most strenuous test any system vendor performs (except DEC, who
pours burning oil into a chassis and makes sure that is not burning
by the time it leaks out the bottom.  Really.  I mean it.)

Nobody can guarantee that a system won't ever have problems.  We can,
however, do our best in engineering these systems to make them as
bulletproof as possible.

5 HW/SW + 2 Tech Support/Assembly

Yes, yes, no (so far, kind of pokey), and yes.

A comment that I'd like to make is that Deskstation, unlike any of
the RISC clone manufacturers, holds our destiny in our own hands.
This is what allows us to get to market with a 21164/300MHz design
(that roughly doubles the performance of the current generation
Alpha) months before our competition.  We don't have to wait
for NT 3.51 because we have engineers that can write BIOSes and HALs
that support NT 3.5.  Nor do we have to wait for chipsets or
reference designs.  If DEC decided tomorrow to stop selling support
chipsets for their processors (DEC has never even hinted this), then
we'd be the only ones who could keep moving forward.

>>The next generation of Alpha chips from DEC are supposed to run in the 600Mhz
>>to 900Mhz range (not exactly sure).  So what specifically is it about the
>>design of the Raptor 3 that guarantees that CPU modules can continue to
>>be developed for it, and just simply plug in?
>
>Semiconductor technology changes as rapidly as any other part of the PC
>environment. Who can predict what types of board interfaces and designs will
>be necessary for the future processors, cache modules, ASICs, etc.

Nobody can predict all of these things in a crystal ball, but it's not too
hard ro predict some of these things if you sign the proper
Non-Disclosure agreements :-)  In the next few years, unless RAMBus
stages a major coup, processor and cache technologies will change
MUCH faster than either memory or backplane technologies.

The part of the Raptor 3 that allows me to assert that it will be a
useful rendererl for about 2 generations of processors beyond the 21164 is that
the CPU-to-cache interface is the precise part of the system that
dominates its performance for compute-intensive tasks (like, for
example, 3D rendering).  That's why we put those parts on the module.
The simulations we run during the cache and memory architecture phase
of a system design are POVRay renders of the standard PacMan scene.

>
>>At what point does the design of the main motherboard itself become obselete?
>
>When new technology renders it obselete. How long is that? Look to the past
>for answers.
>

Or you could ask me => about two processor generations beyond the 21164.

>
>We based Flight Technologies on the idea that any part of the PCs we sell
>would be replaceable to provide upgradeability. This extends to the
>motherboard, but not past it. IMHO, the cost of the electrical parts that
>make up the motherboard are small in comparison to the processor and cache,
>hence the financial gain from keeping a backplane is minimal in the total
>scheme of things, especially when you factor in the rapid pace of innovation
>with bus and memory architectures.

Deskstation, on the other hand, doesn't think that people want to
pull the motherboard out of their systems.  I'm not particularly
comfortable doing that and I design computers for a living.  That's
why the CPU/cache module goes on something that looks like a long PCI
or ISA add-in card.  I'm kicking around ideas for the next generation
of motherboard right now.  Are people comfortable enough changing
motherboards that it would make sense to make future motherboards
backwards compatible with current modules, or would you guys probably
just start over anyway?  To extend the Pentium overdrive analogy,
this would be identical to putting a 486 socket on a Pentium
motherboard, so that you could start by using your old processor and
upgrade later.

Except for the dual SCSI controllers, every
component in a Deskstation system is just as mobile as a component
from a Flight system.  After all, where do you think the Flight guys
learned which devices were cross-platform compatible?

In addition, the ease-of-swap feature allows us to mobilize our
dealer channel to provide an upgrade rebate (of some sort) to the
consumer.  We are incentivizing (probably not a word) our dealer
channel to give an upgrade customer some discount on a new module by
giving that dealer a cut price on a new motherboard, so that he can
move the old module back out to another customer.

Sorry to be so long-winded.  I usually don't read this list, so any
flames or questions (or good jokes) should probably be directed or
copied to [e-mail].



Blaise Fanning
Chief Technology Officer
Deskstation Technology
[phone]
[e-mail]

Blaise Fanning
Chief Technology Officer
Deskstation Technology
[phone]
[e-mail]