Path: utzoo!utgpu!watmath!uunet!ginosko!gem.mps.ohio-state.edu!tut.cis.ohio-state.edu!att!dptg!lzga!jlw
From: jlw@lzga.ATT.COM (J.L.WOOD)
Newsgroups: unix-pc.general
Subject: Re: .5 + .5 + .5 + 1.5 = 2.0 ?
Message-ID: <1680@lzga.ATT.COM>
Date: 25 Oct 89 23:01:32 GMT
References: <4224@itivax.iti.org> <10383@cbnews.ATT.COM> <388@msdrl.UUCP> <1685@mtunb.ATT.COM>
Reply-To: jlw@lzga.UUCP (J.L.WOOD)
Organization: AT&T ISL Lincroft NJ USA
Lines: 17

John McMillan stated that the 7300 would fail with many standard
256Kbit parts.  I have observed this to be true. Unfortunately the
memory diagnostics do not always detect this.  What happened to
me a few years ago was that I was populating 2-EIA + 1.5 Meg expansion
cards with WECo 256KBit parts supposedly industry standard.  The
Combo card tri-states the 8-th address bit as it strobes across
the addresses refreshing the DRAMS.  The WECo chip couldn't handle this
and failed to properly refresh.  The memory diagnostics didn't
fail because their algorithms caused affected memory locations
to be revisited well within the memory refresh interval thus
refreshing whole rows of memory.  I didn't get my failures until I
started loading and running UNIX which will load code into
memory and then revisit it only when necessary.  I got panic traps
up to 10 minutes after reloading UNIX since not all of my memory
on any given machine was of this type.

Joe Wood