Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!rutgers!columbia!cs!dr
From: dr@cs.columbia.edu (David Robinowitz)
Newsgroups: comp.sys.mac
Subject: Re: 16Mhz  why in the world....
Summary: A static memory cache can allow faster processor speeds (or at
	 least get rid of some wait states) w/o having fast MM.
Message-ID: <153@cs.columbia.edu>
Date: 24 Mar 89 03:50:39 GMT
References: <430095@hpcea.CE.HP.COM> <42340b22.a590@mag.engin.umich.edu>
Organization: Columbia University Department of Computer Science
Lines: 46

In article <42340b22.a590@mag.engin.umich.edu>, mystone@sol.engin.umich.edu (Dean Yu) writes:

>   From what I understand, there are a few reasons why Apple hasn't gone with
> a faster clock rate.
> 
>       -  Faster clock rates require faster RAM memory.  And the price of
>         of are ungodly.  If people are complaining about Apple's prices
>         now, think of how much they would charge for a machine with <100ns
>         memory chips.

From what I understand of caches, one could build a static-RAM cache
between the MC68030 and the main memory (slow DRAMS) and get rid of
some wait states/utilize a faster u-processor.  Of course, this would
require sophisticated hardware which would handle
cache-misse/writes/replacement (if [set]associative).  The speedup
would be proportional to the cache-hit ratio which would vary
depending on the application.  I'm no motorola expert, but from what
I've read, the MC68030 has a small data cache and a small instruction
cache onboard.  Enlarging this cache may be all Apple needs to do.  As
my CU professor said, [and I paraphrase badly] "Caches are one of the
few elegant improvements one can make to a computer system.  For a
small increase in $$$$, one can get a significant speedup (more than
just linear)."  Certainly, the MM speed is still important (everything
in the cache has to be read in from MM, and everything that gets
written to MM eventually has to be actually written there) --- I don't
know if 120ns DRAMS are fast enough to support a 33Mhz 030 without
just adding more wait states for MM access.

> Dean Yu                            | E-mail: mystone@{sol,caen}.engin.umich.edu
> University of Michigan             | Real-mail: Dean Yu
> Computer Aided Engineering Network |            2413 Kelsey House
> ===================================|            600 E Madison
> "These are MY opinions." (My       |            Ann Arbor, MI 48109
>  employer doesn't want them.       |===========================================
>  Actually, they don't really care  | 
>  what I think.  But President      |   This space intentionally left blank.  
>  Duderstadt does...)              

It seems Apple is not too concerned w/speedup.

-Dave
Real Men Don't Write Disclaimers.
-- 
----------------------------------------------------------------
David Robinowitz                       dr@cunixb.cc.columbia.edu
CU SEAS CS MS, NYC  			          (212) 601-1586