NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/sys/amiga/misc-part2/10022

Internet Message Format | 1996-08-05 | 10.2 KB

Path: news.vol.it!news
From: [e-mail] (Fabio Bizzetti)
Newsgroups: comp.sys.amiga.programmer,comp.sys.amiga.games,alt.sys.amiga.demos,comp.sys.amiga.misc
Subject: Re: AB3D II beats Quake....
Date: 26 Mar 1996 22:49:28 GMT
Organization: Video On Line
Distribution: world
Message-ID: <[e-mail]>
References: <74000105753944194756@BIRDLAND>
NNTP-Posting-Host: molcl2.vol.it
X-Newsreader: THOR 2.22 (Amiga;TCP/IP)


> bizzetti%mbox.vol.it@INTERNET wrote :

>> PC = only HighEnd CPU, standard chips.
>> Amiga = cheap CPU, clever custom chips.

>Well, maybe that "chip CPU" thought was the problem ?

Today, in modern computers most of the CPU time is used for graphic purposes.
Why do you think that PC people needs a 200Mhz PentiumPro? For games, to handle
what the SVGA can't handle (lotsa things), for OS what Windows95 is slow to
handle. We already have a good OS, now it's time to get hardware custom
solutions to remove all the silly and heavy needless gfx computations from
the CPU. An example from the past? Why the Amiga was much faster than ST in
scroll? (BESIDES hardware scroll): because the CPU didn't have a fast bit
shifter, and thus a *simple* custom circuit has been included in the blitter
for this job. This was versatile enough to serve both games and OS (rastports).
I dont wish 3D boards: they're specialized and destined to die as soon as
the tastes change. What I wish is to study the CPU, and see in what kind of
compuations it lacks, thus "integrate" these facilities into a simple and
cheap hardware, expandable too. But a single family, or all the advantages
will be lost in this hopeless and silly "abstraction of hardware" theory.

Do you think that F1GP-II would still need a fast Pentium if it had a "decent"
video chipset?

"Decent" doesn't mean "10 years to design" and "10000000000$ dollars to make",
it means a solution designed by who have to use it: both assembly realtime
game coders and OS programmers.

If c2p was obligatory ( = not possible to handle in hardware ), to perform it
would you prefer a PPC604, or only a "1000 transistors" simple Akiko?
Are we stupid?
I bet "Akiko" style solution would be still faster, and extremely cheaper.

Just a simple solution like the one I proposed (AgaEXTENDER), that has to
fight with the unilateral structure AGA->RGB, already allows stuffs that no
CPU can make in realtime, and, also if the most expensive CPU could, would you
prefer to spend 3000$ for a CPU board and go with it or 30$ for a custom IC?


>> And dont think that PPC603 is that fast.. Pentium Pro is much better.
>> And anyway, what about older Amiga users?

>Well, did not AT want to use the PPC604 ? As to the numbers i saw Pentium Pro
> did not look very good against PPC604...

At the time that the high-end AmigaPPC604 (although they said 603 for mid-end)
arrives in 1997, I would compare the 604 with the 80686 that will be already
*old*, and maybe in multiprocessing boards as standard for most tomorrow's PC.

While the PPC620 (if not cancelled because, as IBM said, "it's not much faster
than PPC604e") will be a 7.5 millions of (relatively) wasted transistors, if it
can't outperform PPC604e so much. Which more improvements will the PPC630 have?
At IBM they ran out of brain.

I see again a future for CISC: as the most expert of you know, the RISC haven't
been invented yesterday, they are old as me. The RISC vs CISC "war" has always
been combacted with RAM speed as main arm: when the RAM were relatively slow,
the CISC was faster than RISC; when the RAM technology was faster than CPU's
one, the RISC's were faster than CISC's. CPU technology has grown up a lot
last 10 years, while RAM technology has grown up much less. Some years ago the
on chip caches "memory" became so advanced that they've been able to mask RAM
speed a lot, and RISC became more powerful than CISC also in the moment more
favourable to CISC (RAM were slow), so everyone have seen a future for RISC.
IMO, the best move has been made by Acorn, with the ARM, excellent as "idea",
but chocked by Acorns' poor CMOS 5Mhz technology (but we still remember how
it was faster than A500's 7Mhz lotsa transistors CPU), but now things will go
again at CISC advantage. The caches have a limit (already reached) that make
become any improvement as useless. Then the RISC technology had another
advantage: the simplicity. The bug in the firsts Pentiums' FPU showed an
example of the problem, but gave a false idea. CISC has turned page: Pentium
has reached the limit of its technology (meaning that it can't go faster than
this (at the same Mhz), but they went into massive parallel processing, and
the 80686 will allow state of the art multiprocessing capability, meaning that
having a Pentium already "debugged", they'll only have to put 8 into a chip,
and it'll work greatly, gaining a lot from its old 8bit (YEAH) instruction set.
Advantages? 80x86 can contain upto 4 instruction codes into 32bit, while the
PowerPC can contain only 1. This means that parallelization will allow 80x86
to run 4 times faster than the fastest of PPC.

We'll get soon BiCMOS technology for CPU's: 700Mhz, while the RAM will run at
a speed hugely inferior. That day (in 1-2 years) having more concentrate
programs (80x86 = upto 8 instructions every 64bit / PowerPC = upto 2) and
being able to perform more things with each instruction ( = CISC philosophy)
will outperform RISC's of lotsa times.

Intel is not dumb, they said 3 years ago what I understood nowadays.
Time for other people to understand it as well.

The only thing that can allow a future to RISC is ~5ns RAM, which is unlikely
to happen, indeed.

So, we are back again.
The AmigaPPC604 (note: expensive high-end model) is not standard yet, while the
200Mhz PentiumPro is already available and going to be surpassed soon by the
new much faster 80686/80786 "lotsa-Pentiums-into-a-chip" processors.

Of course, being the 680x0 abandoned, I can accept the PowerPC choice without
being worried *ONLY* if we help it in all the tasks that would waste all its
precious CPU time.

Having a separate Audio/Video subsystem, and any CPU (PowerPC is the best
choice due to its commercial quality: but it *cannot* do all alone or it
will become the *worst* choice).


Lesson terminated. ;)

>> IMO a device like the one I designed (AgaEXTENDER) is the best and only
>> solution for all Amigas out there, and giving some feedback from AT
>> engineers it could become perfect as much as possible and needed.

>Well, tell us what comes out of it (but with the Amiga Walker i think GFX
>Boards can be a solution for a BROADER public too...

Nope.
And it would not be for the A1200 though..
When the Picasso comes for A1200, we'll have to laugh. First: unless you've a
200Mhz CPU, you can't expect on Amiga the gfx performances of 200Mhz PC just
"stealing its gfx chip". You forgot to steal the 200Mhz Pentium CPU as well ;)
Then, we'll laugh twice when we'll connect the Picasso to the slow PCMCIA port.

Still not got the idea? I am extremely worried.. =(

>as it has slots...

How many $$$ will low-end people want to invest (to get nothing btw)?

>but we  will see...)

Yea, wise statement.
Anyway, I would prefer to not see tomorrow the death of the Amiga, and see
today the planning of a *GREAT* (read: PC killer) future, since it's *possible*
if you listen to Amiga experts instead of PC/API/C++/virtualization "experts".

Trying to not offend anyone: it's the truth.

>Anyway for the first glance at your text document, it looked interesting..

It has many more hidden features than visible ;)
All made through the agaextender.library (that is *NOT* an API), thus allowing
full future hardware compatibility (like PSX) with *NO* loss of performances.
Useless to say 2 things:

1) It's a low-level extremely optimized OS callable but independent library.
2) You can write an agaextender.library completely optimized for any SVGA
   GfxBoard and get your precious compatibility, but it'll be damn slow,
   showing how lame standard chips are at the Video functions.
Still fully OS compatible though.

The agaextender.library (already designed; if AT wants it to become reality,
it has only to give the start) is meant as a clever interface: it accepts
a binary "program" (into a special language, that can be translated from
mnemonics to binary or directly generated by a routine) and sets up OR
(important) simply updates everything on both AGA and AgaEXTENDER registers
to get the wanted result. It can work on a horizontal pixel basis, meaning
OS horizontally draggable screens too if you want (just to let you understand
what performances I am talking about). Changing a part of the program will not
mean that agaextender.library will "recompile" the whole program, but only
the parts that changed. The LAMEST example would be to tell it (in a simple
way) to generate (raster-realtime) a screen where 2 hardware screens are
present: one mid resolution in YUV+"I"decompression, one with very high
resolution for text, all not in a vertical basis (as AGA), but also horizontal:
meaning PIP (picture in picture).

The CPU would only need to write in the 2 zones of the RAM, without the need to
call the library anymore, and all would be magically working on the screen.
If you dont like things in realtime, you can still write a software version
of agaextender.library for SVGA cards.

> i hope you can do "Standard" resolutions with it too...
>something like 320x256 Chunky... (in the text you mentioned only "up to xxx x
> yyy" with xxx x yyy being some strange resolutions... but i will have to do
>a  more careful look at the Text...


You can have also 321.99*1111, or 999.01*100. And it doesn't require a separate
programmable clock generator to allow *any* resolution: it uses just a clever
and simple way to do it (better if you examine the docs).

>Steffen Haeuser

I am doing all I can (humbly, and talking only when I can bet my life on the
things I say) to save the Amiga from a future of low-end/mid-end death, and
poor and limited high-end short future.

  /-----------------------------------------------------------------------\
  |  Fabio "Maverick" Bizzetti - [e-mail] - Maverick* at IRC  |
  |             The maker of "CyberMan" and "Virtual Karting"             |
  |               working on "VirtualRally" & "StarFighter"               |
  \-----------------------------------------------------------------------/