Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uwm.edu!zaphod.mps.ohio-state.edu!usc!snorkelwacker!bloom-beacon!eru!luth!sunic!mcsun!ukc!acorn!moncam!emmo
From: emmo@moncam.co.uk (Dave Emmerson)
Newsgroups: comp.sys.ibm.pc
Subject: Re: Memory Problem: How Best to Diagnose?
Summary: TC++ as a memory test?
Message-ID: <683@marvin.moncam.co.uk>
Date: 6 Jul 90 03:54:42 GMT
References: <1990Jun28.032323.5496@ux1.cso.uiuc.edu>
Organization: Monotype ADG, Cambridge, UK
Lines: 26


Sounds like your memory tests are a bit simplistic -
fill a huge area with known or predictable data, then
read back and compare.

Trouble with this type of test is that REFRESH problems
are disguised by the test itself.

Typical refresh reqirements are (for each chip) 512
consecutive addresses to be accessed every 8msec.
Your memory tests are exceeding that by a HUGE margin
by thundering through your DRAM at top speed. If you
decide to 'roll your own', do the fill first, then
WAIT 2 or 3 seconds in a loop in low memory before
you read back and compare. That way, if the data's
volatile, you stand a chance of detecting it.
Natch, you should avoid haveing any other process
using the ex??nded memory during the test, so you
may need to alter your config.sys..

Good luck!

Dave E.

NB: I assume the SIMMs/SIPs meet the minimum setup time
    for your system- you did check didn't you?