Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!shadooby!samsung!usc!rutgers!cbmvax!daveh
From: daveh@cbmvax.UUCP (Dave Haynie)
Newsgroups: comp.sys.amiga
Subject: Re: Kronos vs. Hardframe
Message-ID: <8687@cbmvax.UUCP>
Date: 22 Nov 89 22:43:07 GMT
References: <3302@convex.UUCP>
Organization: Commodore Technology, West Chester, PA
Lines: 52

in article <3302@convex.UUCP>, swarren@eugene.uucp (Steve Warren) says:
> Keywords: Kronos, Hardframe, controller

> The reason the Kronos can beat a DMA controller is because it has a 16-bit
> path to memory.  Other controllers only have an 8-bit path to memory.  

Pure bunk.  All DMA controllers have 16 bit data paths.  You'd have to go 
to a great deal of extra trouble to build a DMA device that supports 8 bit
transfers.  None of the Commodore controllers even support 8 bit 
transfers, and I suspect the other don't either -- there's never any
reason for that in a DMA controller.  Among the non-DMA controllers,
you're split.  The GVP controller has a 16 bit data path, I believe.
So does the C Ltd. Kronos.  The low cost TrumpCard from IVS and the
old C Ltd. controllers also have 8 bit data paths.

> I think that they have a significant performance improvement over other
> controllers because of the path-width advantage.  

The only possible advantage a Kronos card is likely to have over any
other 16 bit card is in the software.  All the Commodore cards have very
tightly written assembly coded device drivers.  So does the Hardframe,
far as I've heard.  The GVP's controller is written in C language and
suffers a bit because of it.  I'm not sure about the others.

> Their literature is misleading...

If they've managed to make you think that DMA controllers are doing 8 bit
transfers, they're far more misleading than you imagine.

> claiming that DMA controllers are fundamentally inferior when there is 
> chip-ram contention.  This is a false claim, and it is also unnecessary 
> if the disk-perf results they published are true. (Their disk-perf results 
> were significantly better than all the others)

What were their DiskPerf results?  Our disk folks are seeing over 1 meg/sec
with 2091s through the filesystem, with a good hard disk.  The main trouble
I've seen with the magazine reviews of HD controllers is that the reviewers
don't quite understand or bother to deal with a real, sound, scientifically
accurate comparison.  For such a test, you've got to benchmark each controller
with exactly the same hard disk, and preferably several different ones, under
a variety of system load conditions.  I haven't designed a hard disk
controller myself, but unless there's something really stupid in the design
of the 2091 and the Hardframe that I've somehow missed up to this point,
there's no way in the world any CPU driven controller should benchmark
faster than either of there in a fair test on a stock A2000 or even a
stock A2500/30.

> --Steve
-- 
Dave Haynie Commodore-Amiga (Systems Engineering) "The Crew That Never Rests"
   {uunet|pyramid|rutgers}!cbmvax!daveh      PLINK: hazy     BIX: hazy
                    Too much of everything is just enough