Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!mips!dimacs.rutgers.edu!rutgers!cbmvax!daveh
From: daveh@cbmvax.commodore.com (Dave Haynie)
Newsgroups: comp.sys.amiga.advocacy
Subject: Re: NeXT software size
Message-ID: <21529@cbmvax.commodore.com>
Date: 13 May 91 21:36:34 GMT
References: <48808@ut-emx.uucp> <-h6Hqu=@cs.psu.edu> <48816@ut-emx.uucp>
Reply-To: daveh@cbmvax.commodore.com (Dave Haynie)
Organization: Commodore, West Chester, PA
Lines: 40

In article <48816@ut-emx.uucp> greg@ccwf.cc.utexas.edu (Greg Harp) writes:
>In article <-h6Hqu=@cs.psu.edu> melling@cs.psu.edu (Michael D Mellinger) writes:

>I'm no EE, either, but I think a 25Mhz 040 using 100ns RAM would be at
>least a one-wait-state configuration.

An '040 at 25MHz will take at least three wait states using a single bank of
100ns DRAM.  But you don't normally speak of '040s running single cycles,
anyway, because their normal mode of memory access is similar to the '030's
"burst" cycle, for both reads and writes.  Those three waits would be 
taken on the first bus cycle of the burst.  The next three cycles could very
likely be one clock each, if the system uses nybble-mode DRAM, or two clocks
each, using page or static column memory and some external logic.  With 80ns
DRAM on an '040 bus, you can just manage two wait states, as it has a 
somewhat more efficient bus cycle than the '030.  With page or static column
memories, you can have memory banks as small as 1MB.  With nybble modes, you
typically need 4MB banks.  Both access methods may lend themselves to various
bank interleave schemes, which can up the performance using the same kind of
memory, at the cost of more external logic and a larger minimum bank size.
Minimum bank sizes can also vary by device type -- some of the newer 16 bit
wide DRAM can cut this down for you, though standard SIMMs are all 8 bits
wide.

>Anyway, I don't understand what makes using slower RAMs that much cheaper.
>The memories themselves woudn't be much different in price in that kind of
>volume.  Is it an architecture thing?  Anyone?  Anyone?  

They chip yield is the main factor.  If 99% of the 80ns dice work, but only
40% of the 70ns dice, you'll see a lower price for 80ns parts.  It may also
reflect a more expensive chip fab on the faster part, or just market realitites
(eg, they can get it because there's not much competition).  Currently, you
pay about the same for anything 80ns or faster, which implies that most of the
chips made work at 80ns.  50-70ns parts are still something of a premium.

>       Greg Harp       |"I was there to match my intellect on national TV,

-- 
Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests"
   {uunet|pyramid|rutgers}!cbmvax!daveh      PLINK: hazy     BIX: hazy
      "That's me in the corner, that's me in the spotlight" -R.E.M.