Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!uwm.edu!cs.utexas.edu!ginosko!uunet!cbmvax!daveh
From: daveh@cbmvax.UUCP (Dave Haynie)
Newsgroups: comp.sys.amiga
Subject: Re: 68030 Accelerators, Info Wanted
Message-ID: <8357@cbmvax.UUCP>
Date: 30 Oct 89 17:57:56 GMT
References: <14325@well.UUCP>
Organization: Commodore Technology, West Chester, PA
Lines: 57

in article <14325@well.UUCP>, theobaby@well.UUCP (Paul Theodoropoulos) says:

> Dave Haynie of C-A writes [regarding the Ronin 030 board] -
>>First of all, an MMU table of any kind is going to waste some memory.
> Well, i suppose so, though i can't imagine too many folks with 030 systems
> having memory constraints so severe that they can't afford a coupla dozen
> kilobytes for some page tables.

No problem there.  Only, the software only method Ronin uses doesn't allow
caching of 16 bit RAM.  If you're only using 32 bit RAM, that may not be a
problem, but it's certainly a disadvantage.  Plus, the software-only method
will either result in lower performance or more memory wasted on tables than
a hardware/software combination.  Because you need separate tables for I
and D cache inhibit if you're software-only, or you end up inhibiting the
I-cache were it's totally unnecessary to do so (in CHIP RAM, expansion RAM,
etc).  This is a complaint, not the end of the world.

> Secondly, can you explain why the Ronin board tends to do better on almost
> all benchmarks against other, equivalently clocked 030 systems (such as the
> Commodore 030)? 

Probably because it doesn't.  First of all, the Ronin '030 hasn't been compared
to the Commodore '030, at least anywhere I've seen.  None of the magazines have
a Commodore '030 board, and I sure don't have a Ronin '030 board.  The Ronin 
board does use a 28MHz clock, rather than the 25MHz clock used by Commodore
and GVP.  I think both Commodore and GVP designs work just fine at 28MHz (I
run my office board at 28MHz) without changing anything but the CPU (Motorola
CPUs come in 16.667MHz, 20.000MHz, 25.000MHz, 33.3333MHz, 40.000MHz, and
50.000MHz speeds -- running a 25MHz part at 28MHz, while you might get away
with it most of the time, is "asking for it").

> Seems to me that that vindicates their use of software memory control.

No.  All software memory control does for you is simplify hardware design by a 
very small amount, and of course _require_ the use of the software.  The use
of an MMU will in fact slow the system slightly, since every time that CPU hits
a page that's not in the ATC (22 fully associative entries in the '030), the 
MMU has to go out and find the new page.  So to the purist at least, the software
only method is guaranteed to be slower than the hardware only menthods.  In
reality, you probably won't notice the difference, especially if you use short
trees in your MMU software -- SetCPU tables never get more than two levels
deep, and the second level is only for supporting expansion ROM translations
(since expansion ROMs are small).  And of course, software is software -- any
benefit available on a Ronin board under software control is available on
a Commodore or GVP board.  

Like I said, I don't think there's been a real comparison of real '030 boards.
I know some of the magazines are waiting for Ronin and Commodore boards to do
just such a comparison.  But whatever the results, be assured that you can 
look elsewhere to explain any performance differences (the 32 bit memory system
is always the first place to look).

> *****Stamp out bandwitdth wasting signatures! Paul Theodoropoulos*****
-- 
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