Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!cbmvax!daveh
From: daveh@cbmvax.commodore.com (Dave Haynie)
Newsgroups: comp.sys.amiga.advocacy
Subject: Re: Amiga bashing
Message-ID: <22762@cbmvax.commodore.com>
Date: 28 Jun 91 04:42:05 GMT
References: <1991Jun11.204407.16603@murdoch.acc.Virginia.EDU> <1991Jun20.200326.16487@bmerh409.bnr.ca> <1991Jun21.002542.19989@mintaka.lcs.mit.edu> <1151@stewart.UUCP>
Reply-To: daveh@cbmvax.commodore.com (Dave Haynie)
Organization: Commodore, West Chester, PA
Lines: 54

In article <1151@stewart.UUCP> jerry@stewart.UUCP (Jerry Shekhel) writes:
>rjc@wookumz.gnu.ai.mit.edu (Ray Cromwell) writes:

>>A $2000 386 system probably doesn't have decicated DMA-I/O
>>(most likely it is using polled I/O which incurs a major performance
>>hit on a multitasking multithreaded file system).

>Wrong.  Every PC has DMA I/O.  

Actually, you're both wrong.  Or right.  PCs have do have a DMA controller, 
but that's nothing similar to Amiga DMA.  Bascially, when the PC first came
on the scene, IBM noticed what a bad job the 8088 did at block copies of
memory.  So they put in this DMA controller thing, which also does their
DRAM refresh (they got a patent on that part of it).  Anyway, you can use this
for memory copies or for memory to/from I/O.  That's the good new.  The bad
news is, any decent processor will do the job better anyway.  Any Motorola
processor from the '010 on can do block memory copies without the need to
fetch instructions.  So this kind of DMA controller is pretty much useless in
a modern system.  Amiga hard disk DMA, on the other hand, is a big win.  Since
the Amiga DMA controller sits between the memory and the I/O device, it only
takes one bus crossing per word transferred.  IBM style DMA controllers, and
CPUs in block move mode, take two bus crossings per word transferred.  So the
Amiga's DMA is, without even considering it's fancy features like the FIFOs we
use in A209x and A3000, twice the speed of any IBM style DMA controller.  In
the PC world, though, the news gets worse.  The speed of the DMA controller in
these systems hasn't increased with CPU speed.  So, if you're on a basic AT on
up to a butt-kicking '486 system, you lose big if you use the DMA controller.
So most of these guys use programmed I/O.  Not polled, they do have interrupts,
though with disk transfers from unbuffered media, interrupts are inefficient.
So there's really no decent disk solution on the basic PC Clone.  You need an
add-on buffered bus-mastering SCSI subsystem on MCA or EISA bus, or something
similar, to start looking like an Amiga rather than a Mac in the disk 
department.

>I run UNIX on the cheapest 386 possible, and intense disk I/O going on in the 
>background does not slow down the machine significantly at all.  

Of course it does.  Fast motherboard EDSI on any PClone can take more than 
50% of your CPU time away from you, if you have constant disk activity.  That
same disk activity on the A3000 will amount to a loss of around 5% of your
CPU time.  But we're working on it, we'll probably see 2.5-3% in the next 
generation :-)...

>Maybe that's because there have been so many improvements in VGA cards
>recently, that prices have been going down?  

Also, these new Taiwanese 15 bit CLUTS have reduced the price significantly
at the low end.  RAMDAC prices go up pretty steeply as you add increase the
palette.

-- 
Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests"
   {uunet|pyramid|rutgers}!cbmvax!daveh      PLINK: hazy     BIX: hazy
	"This is my mistake.  Let me make it good." -R.E.M.