At 15:30 -0400 08/13/2005, PCI PowerMacs wrote:
Date: Sat, 13 Aug 2005 11:08:45 +1000
From: David Elmo <[EMAIL PROTECTED]>
Today, 15 starts. But Gauge Pro in a spare moment when I have a break, has
come up with a memory error, which might explain all this nonsense that has
been going on for so long. "Memory Error at address $01FF4AB4 but read
$05FF4AB4" I don't suppose anyone can say how i can usefully use this notice
to pinpoint the chip concerned? If the following gets all unwrapped, I am
sorry.
Do you remember how many iterations before the failure appeared?
Some memory defects are obvious (memory cell stuck at 1 or stuck at
0) and some are subtle (requires certain pattern of data in
surrounding cells to trigger problem in defective cell). The obvious
defects are usually detected within a few score iterations. The
subtle defects require up to 1300 iterations to detect.
To reliably test your memory, you need to use the RAM Sandwich
method. This involves testing with three DIMMs at a time. Put known
good DIMMs in A1 and A4 (or B1 and B4) and put a DIMM to test in A2
or A3 (or B2 or B3). Why this arcane configuration?
When the memory testing software is running it can't test the memory
occupied by the Operating System. So even if the tests pass, you
don't know that your RAM is good because there may be a defect in the
portion of RAM which the OS occupied. We're not sure if A1 or A4
(and B1 or B4) corresponds to the lower memory addresses, so we need
known good memory at each end, plus there are indications that the OS
uses some memory at both the top and the bottom so either way, you
need a "filler" DIMM on each side of the DIMM being tested.
If you don't have any known good DIMMs you can still run this test.
First, test your DIMMs one at a time. This will quickly week out any
severely defective DIMMs. A one-at-a-time test has the following
characteristics. A failure is always a failure. A Pass is never
reliably a Pass. In other words, if the DIMM fails, you know there
was a defect somewhere (absent defective DIMM sockets, or overheated
CPU). If the DIMM passes, you don't know for sure that it is good
because there could be a defect in the portion that is home to the OS.
Next use the RAM sandwich method. If you like you can put two DIMMs
in the middle in both A2 and A3. In your case, I'd start off with
64 MB DIMMs in A1 and A4, just because lower capacity RAM will take
less time to test. The RAM Sandwich method with "unknown bread" has
the following characteristics. A Pass is only a Pass for the middle
DIMM(s), i.e. the meat of the sandwich. A Failure doesn't tell you
which of any of the DIMMs (meat or bread) is bad. Only that one or
more of them is defective.
So a Pass with the RAM Sandwich method using "bread" of unknown
quality tells you that the "meat" is defect-free. That is how you
will find your known good DIMMs to use as bread in later tests.
Once you've had one or two successes with the Sandwich method, you'll
have your two known good DIMMs. Then you can use those for the bread
and test the rest of your DIMMs. A RAM Sandwich with known good
"bread" has the following characteristics. A Pass is a Pass for the
meat of the sandwich. A Failure is a failure for the meat, because
you already know that the bread is defect-free. If you're testing
two DIMMs as the meat, then on a failure you'll need to run another
test to figure out which of the two is the bad one.
With RAMometer or Gauge Pro, you need to run at least 1300 iterations
to be fairly certain that your RAM is defect free. How did I arrive
at that number? Over on the SuperMac List 1200 and something is the
highest number at which anyone found a consistent failure. There
might be failures that require more iterations to detect, but we
haven't run into them yet. However, we *know* that some types of
defect require over 1200 iterations to detect and occur consistently
on the same iteration over 1200.
Credit for development of this method should go in part to Velocity
Upgrades who shipped so much defective RAM, we had to develop a
reliable methodology to test their product.
Jeff Walther
--
PCI-PowerMacs is sponsored by <http://lowendmac.com/> and...
Small Dog Electronics http://www.smalldog.com | Refurbished Drives |
-- Sonnet & PowerLogix Upgrades - start at $169 | & CDRWs on Sale! |
Support Low End Mac <http://lowendmac.com/lists/support.html>
PCI-PowerMacs list info: <http://lowendmac.com/lists/pci-powermacs.shtml>
--> AOL users, remove "mailto:";
Send list messages to: <mailto:[email protected]>
To unsubscribe, email: <mailto:[EMAIL PROTECTED]>
For digest mode, email: <mailto:[EMAIL PROTECTED]>
Subscription questions: <mailto:[EMAIL PROTECTED]>
Archive:<http://www.mail-archive.com/pci-powermacs%40mail.maclaunch.com/>
iPod Accessories for Less
at 1-800-iPOD.COM
Fast Delivery, Low Price, Good Deal
www.1800ipod.com
