Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!usc!apple!uokmax!d.cs.okstate.edu!ong
From: ong@d.cs.okstate.edu (ONG ENG TENG)
Newsgroups: comp.sys.ibm.pc.hardware
Subject: Re: Mixing 1M and 256K simms- Why not?
Message-ID: <1991Feb24.001345.6220@d.cs.okstate.edu>
Date: 24 Feb 91 00:13:45 GMT
References: <1991Feb23.045704.985@cbnewsd.att.com>
Organization: Oklahoma State University
Lines: 42

From article <1991Feb23.045704.985@cbnewsd.att.com>, by bash@cbnewsd.att.com (thomas.w.basham):
> 
> There seems to be a consensus (or myth) that a system using SIMM
> memory can't mix 1M and 256K simms.  My machine uses two banks of
> 4 slots.  Bank 0 must be filled, bank 1 is optional.  Since the
> system obviously has the ability to determine bank 1's status,
> why is it beyond reason to simply see how high one can access bank 1,
> add that to what's in bank 0, and store it in CMOS.  
> 
> Those who have 1M in 256K's and upgrade to 4M wouldn't have to sell
> off their 256K's, but now have 5M.
> 
> My question is:  why wouldn't/isn't this done?  Are there valid
> reasons (other than programming shortcuts) to keep memory to a
> 1/2/4/8 M system.
> 
> Or:
> 
> If this mixing can be done, why is so few people seem to know about this?
> 
> Seems to me it's nothing but poor BIOS programming that can't access
> the separate banks with different ranges.  I've mixed ram types in
> (homemade) micros before, albiet (considerably) smaller systems.  It's
> really just a matter of knowing what the top address is, and not
> exceeding it.

Well, sure, you can mix, but you might have to sacrifice CPU power.  What
do I mean?  Well on a 12MHz 286, that's fine, because 80ns RAM's are common
and we can achieve 0 wait state w/o interleaving.  But when you go 
to say 16MHz or 20MHz, 80ns is not fast enough to keep pace with the
CPU, so bank interleave is required for 0 wait state (else you have to
go 1 wait state on NO interleave).  What does interleave requires?
Well instead of having just one bank (that is, 18 bits for 286 and 386SX,
36 bits for 386DX) of the same type of memory (that is 256k chips or 1meg 
chips), you need 2 banks of the same kind.  Since most motherboards
regardless of 386DX or 386SX have only 2 banks on board, mixing memory
chip types mean sacrificing interleaving, i.e. CPU power. 

Yes, I do have 1meg of 256k memory chips sitting around after upgrading to
4meg.  Memories are so cheap today that I can play smash-the-bug with
my hammer on the chips. (They should make a law that requires all 
386's onwards to have at least 4meg)!