Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!uunet!cbmvax!daveh
From: daveh@cbmvax.commodore.com (Dave Haynie)
Newsgroups: comp.sys.amiga.hardware
Subject: Re: Blitter Speed(?)
Message-ID: <15343@cbmvax.commodore.com>
Date: 23 Oct 90 16:28:07 GMT
References: <1990Oct22.195001.28782@idt.unit.no>
Reply-To: daveh@cbmvax.commodore.com (Dave Haynie)
Organization: Commodore, West Chester, PA
Lines: 74

In article <1990Oct22.195001.28782@idt.unit.no> daglem@solan7.solan.unit.no (Dag Lem) writes:

>Would it be theoretically possible to make new custom chips that run at, say, 
>28MHz cycles, and to make the chip-RAM follow this speed? 

Here we go again...  What, exactly, do you mean by "28MHz speed".  One cycle
every clock period?  Every two clock periods, like the 68030?  Every three,
like the 68020?  Every four, like the 68000?  Anyway, the current Agnus 
technology is going about as fast as it will, thanks to NMOS.  Of course,
Commodore and the rest of the world are doing designs in CMOS these days, which
are larger for the same design rules, but run cool enough to go fast and get
large.  Theoretically, a redesigned Agnus chip could run faster.

>Could the pipelining in the blitter be improved to make it blit one word in 
>two cycles, instead of todays four cycles (simple blits)? 

DRAM can't easily support that kind of speed.  Blits take place between 
arbitrary locations in memory.  One memory cycle every clock cycle would 
require memory with a 35ns cycle time.  The only thing in that ballpark is
static RAM, which is way expensive.  If you run some kind of "burst" cycle,
like the 68030, you might manage one long cycle followed by a number of 35ns
cycles that take advantage of nybble, page mode, or static column DRAM.  In
a sense, any speedup trick you can think of that's used by CPUs might also
be adapted to the Amiga chips.  But there are always tradeoffs in complexity
vs. performance.  And consider that clocking isn't the only way to improve
bandwidth -- doubling the bus width is just as good as doubling the clock 
speed for any CPU-type device.

>I ask these questions because I've written routines in assembly using the 
>blitter to draw filled polygons, and it P*SSED ME OFF to see the blitter 
>using one frame just to clear four bitplanes. A 68040 running at 7MHz would 
>do this just as fast, no? Heck maybe even the 68030 in the Amiga 3000 can 
>compete with the blitter, having 32 bit access to chip memory.

As long as you don't have to worry about bit alignments, the 68030 in the
A3000 will probably outpace the blitter.  Clearing a bitplane is skewed far
in the 68030's direction, since the only bit-alignment conditions are at the
bitmap boundaries.  Even when you're dealing with a true non-byte aligned
move, like an arbitrary blit of an image, the 68030 might do it faster, 
especially on the A3000 where the Chip RAM is 32 bits wide.  That's not a 
great surprise, really -- the 68030 is a good, fast, and somewhat expensive
general purpose CPU.  The main points of the blitter are [a] to move bitmaps
around quickly for a very small amount of money and [b] to do this IN PARALLEL
with the host CPU.  If all you're concerned about is a screen clear, a 68030
may do it considerably faster.  But on an Amiga you're never only concerned
with graphics throughput, you're concerned with total system throughput.  Also,
you don't have 68030s yet cheap enough to dump into an A500, whereas the 
blitter functions are essentially free (eg, you need SOME kind of answer to
the basic A500 graphics, floppy, serial, etc. needs, and the Amiga chipset at
this point fits the system better and costs less than alternatives without any
special functions).

>If the blitter could run at 28MHz, 2 cycles per word (8 times faster), we 
>could all forget about other computers. 

What other computers?  Seriously, no matter how you slice it, 

	blitter + 68030 > 68030

>Is it possible???? Anybody??? Dave Haynie?????

What, me again!?  I'm not the chip designer here.  And you don't just crank
out full custom silicon because "it would be nice", there are lots of issues.
And in general, any good design must be carefully architected to fit what
you're plugging it in to.  Simply cranking up a clock rate, whether it's custom
chip or CPU, may sound like a great idea.  But if that 8 times performance
boot costs you 80 times the money, it's probably a bad idea.  You may find an
alternate architecture that gives you 8 times the performance for 8 times
the money, or 4 times the performance for the same cost.  There is NEVER one
simple answer.
-- 
Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests"
   {uunet|pyramid|rutgers}!cbmvax!daveh      PLINK: hazy     BIX: hazy
	Standing on the shoulders of giants leaves me cold	-REM