NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/sys/amiga/hardware-part2/10389

Text File | 1996-08-05 | 8.4 KB | 164 lines

Newsgroups: comp.sys.amiga.hardware
Path: netnews.upenn.edu!dsinc!scala!news
From: [e-mail] (Dave Haynie)
Subject: Re: What can we do with CHRP?
Sender: [e-mail] (Usenet administrator)
Message-ID: <[e-mail]>
Date: Tue, 16 Apr 1996 21:47:32 GMT
Reply-To: [e-mail] (Dave Haynie)
References: <[e-mail]> <[e-mail]> <4kfb6e$[e-mail]> <[e-mail]>
Nntp-Posting-Host: gator
Organization: Scala Computer Television, US Research Center

In <[e-mail]>, [e-mail] writes:
>In article <4kfb6e$[e-mail]>, [e-mail] (Manuel Jesus Jr. ) writes:

>>>While i have everyones attention, has anyone read Dvorak's column in
>>>the April 23rd issue of PC Magazine? If he's right it does not bode
>>>well for the amiga.

>>  Since I don't read PC Magazine, could you please tell us what Dvorak
>> is ranting about this time?

>Well, First of all I was wrong: It is the Dvorak column in the April
>9th issue. Inside Trak. In short he rambles about the MMX extensions
>that Intel is planning to add to the Pentium chips. This is the NSP
>(Native Signal Processing) stuff that has been talked about recently.

Actually, Intel has been talking about NSP quite a bit longer than
MMX. The basic idea of NSP was to add a realtime kernel to PClones,
much like a DSP operating kernel (in fact, Intel was using SPOX, which
is a DSP kernel), underneath Windows. Any realtime stuff that had to
run would get scheduled here, Windows would get any leftover time. It
wasn't such a bad idea, since a kernel like SPOX has been fairly well
beat up in the DSP markets, while Windows even now still doesn't do
realtime very well (Microsoft extensions are coming any day).

Intel's real reason for all of this is simple: you don't need faster
x86 chips to run a word processor. As they find more work for the
existing CPUs to do, the less happy you'll be with today's CPUs (at
least once the novelty of "wow, I didn't know I could do X without
specialized hardware" wears off), and the faster you'll jump on the
next generation. Which, of course, is where Intel gets their money.

MMX is the MultiMedia eXtension to the x86. Simply put, they're adding
some DSP-like instructions to this next generation of Pentium CPUs,
much like Sun did to the SPARC recently.

>Translated to english this would mean that PCs wouldn't need seperate
>sound and video boards because the cpu would be able to do it all.

Not exactly. What it does mean is that the CPU may be able to do more
DSP like things. That means faster MPEG decoding, faster graphics,
maybe some music synthesis, etc. You still need a graphics card and an
audio CODEC. It become possible, at least on the low end, to leave out
a full blow sample-based synthesizer or MPEG decoder, since you can do
these things in software. You don't necessarily WANT to do them in
software even if you can, when the hardware is cheap enough, or when
you have other things in mind for that CPU power. That's not really
what it's about; MME really doesn't displace dedicated hardware except
when price is a real concern, simply because the dedicated hardware
will do the job better, and it won't take a big chunk of your CPU
power away. When this does probably achieve is keeping any
programmable DSP standard off the PClone motherboard. Intel wants
theirs to be the only programmable CPU around. That keeps you needed
faster x86s.

>The only people that would buy such devices would be those that want
>exceptional performance or some other special need.

More than likely, MMX-based x86 processors will replace the older
generation in one pass. This will be something everyone uses, not a
special purpose thing.

>Apparently the PowerPC does a form of NSP right now but the MMX
>extensions would be 4x the speed of the powerpc. This is what dvorak
>claims.

The PowerPC architecture already has more DSP-type operations, such as
MAC instructions, that were not in the x86. MME goes much further, as
does Sun's version of it. The primary enhancements are things like
byte-array instructions, so you can operate on 4 or 8 eight-bit pixels
with a single instruction. These help out in general graphics, and
especially on things like MPEG.

>It is my humble oppinnion that NSP will never fly.

It'll definitely fly, it's over all a good idea. The main reason being
that systems are moving, slowly but surely, to multiprocessing. Now,
I like both homogenous and heterogenous multiprocessing, both have
distinct advantages. The main problem with adding things like
DSP coprocessors (like in the A3000+) is that they tend to be more
costly, especially today, than heterogenous systems, since you're
dealing with bus protocol conversions, etc. And you can't trade DSP
power for CPU power. And you need two sets of development tools. The
advantage of heterogenous multiprocessing is what you've seen in the
Amiga -- if engineered correctly, a specialized processor (like the
Copper) can do a job much better than the general purpose CPU, for
much less cost. This has been used extensively, even on the PC, for
non-programmable processing elements, like blitters and
synthesizers. It's only rarely been done cost effectively with
programmable processing units. The idea of NSP makes the CPU much more
flexible, and as long as homogenous multiprocessing is cheaper, this
will extend that usefulness considerably.

>Why put even more pressure on a cpu when you already need a 100Mhz
>and 8 megs of ram to get exceptible performance out of win95?

Because you're not spending all that much memory on this stuff, you
can get more work with NSP than without it for the same money, 100MHz
this year is 200MHz in a year or two, and who mentioned Windows 95?

>Dvorak seems to think that NSP will be able to carry the PC into the
>desktop video market and other areas that it hasn't been able to get
>into so far.

That hardly requires NSP. Though I think the main point he's missing
is that, unless they've pulled a 180, NSP is basically dead. Intel
announced it, started rounding up developers, met with Microsoft, and
then it went away. Microsoft takes a very dim view of anyone trying to
establish a programming standard they don't control. And it's fairly
clear who's job it is to keep the x86 in the catbird seat. Now, nsp
(no caps) is already going on on any computer system you can
imagine. Microsoft will definitely introduce their own version of it,
and it'll probably be a whole lot worse than Intel's and
simulataneously hailed as What We've All Been Waiting For.

The Amiga's been doing it as long as anyone, thanks to a good designed
in realtime response. And even non-realtime; what do you think
raytracing is, after all. Building a CPU that does this kind of thing
better is not a problem.

>Areas that are the dominion of the Amiga and the Mac. I could be
>wrong but I think Dvorak missed the boat with this prediction. Only
>time will tell.

Clearly, adding some working realtime APIs to Windows will increase
the PC's acceptability in video and audio work; people use it there
already, even though it's not that well suited today. One thing to
notice about the PC business is that, when another computer does
something interesting and new, bad PClone versions of it show up
fairly fast. They eventually get better. Unless there's a HW or OS
wall (which there often is, such as realtime is today), they just keep
getting better. When they're maybe 50%-75% as good as the thing on
another system, that other system had better watch out. Windows is a
good example of this on the PC. So is desktop publishing or desktop
music, formerly the Mac's strong suit. DTV isn't far behind. UNIX
machines are ousted by PC/Windows NT based servers and
workstations.

This is the effect of fairly cheap, multivendor, commodity hardware.
I doubt there's an Amiga fan listening who hasn't wondered "what do I
do if Amiga Technologies fails". Does anyone in the PC business worry
what they'll do with their hardware if OS/2 or even Windows were to
fail, or how they'll upgrade if Compaq or AST went under? Of course
not. Establishing this kind of market is the whole point of the
PowerPC's PPCP standard.

Dave Haynie          | ex-Commodore Engineering |   for DiskSalv 3 &
Sr. Systems Engineer |  Hardwired Media Company | "The Deathbed Vigil"
Scala Inc., US R&D   |    Ki No Kawa Aikido     |     [e-mail]

         "Feeling ... Pretty ... Psyched" -R.E.M.