Hi..
 SM> From working on computers in a repair shop, I can tell you that
 SM> memory problems on the modern motherboards is a total pain in the
 SM> ass.  One has the regular SIMMs, external cache, and internal
 SM> cache, plus the CPU itself, as well as the motherboard/BIOS
 SM> (CMOS settings), each one a potential source of memory problems.
All true.... however you can eliminate most areas of contention easily. 
Remember, i'm testing MEMORY, *NOT* an entire system. The MEMORY is all I'm 
interested in. Using a known good M/B+CPU etc easily gives a platform where 
the memory is the *ONLY* variable in question.
 SM> (in quotes, since most who claim to be technicians are just
 SM> wanna-bes who won't ever make it since they don't care to sit down
 SM> and track down a problem the hard way).
there is a difference between a technician, and a trained monkey. :-)
 SM> I'd say that the hardest thing right now to track down is internal
 SM> cache errors; I've yet to see a program that checks internal
 SM> cache, even though both Intel and AMD Pentium-class chips, as well
 SM> as the most recent 486-class chips (not Cyrix; they don't have the
 SM> extra registers) have special registers specifically for testing
 SM> the internal cache.  Personally, I think it sucks when the best way
 SM> to check for internal cache problems is to swap the CPU and then
 SM> run Win3x/Win95 to see if it acts less flaky.
Well, for starters: internal caches very rarely fail. I've never seen a known 
bad internal cache CPU, ever. It is easy to test them, however, using the 
test registers and/or MSR's. Cyrix CPU's do have the appropriate registers. 
In any case, since the internal cache is software-controllable, making a test 
for it is easy stuff. If the test fails with internal cache on, but not with 
it off, then something related to the cache is at fault.
I have written a cache tester for internal cache, using the MSR/test 
registers (486/pentium/pentium-mmx). It has never yet found a bad CPU. I 
have, however, found plenty of motherboards that have faulty or poorly 
designed bus-mastering support, cache coherency support and/or cache 
writeback designs.
 SM> External cache can be tested simply by disabling it in the CMOS or
 SM> removing the COAST module, and doing the Windows memory test again
 SM> (see commentary about about using Windows as a memory test).
Anyone who uses windows as a system stability test is nuts. Yes, windows does 
test your hardware in a round-about way, but it definately isn't the right 
way to go about things. Finding a good diagnostic tool is the ultimate holy 
grail, which is why I'm writing my own. Disabling external cache is also 
possible under software (regardless of chipset). What is not possible is to 
have one cache on and one off (either they're both on, or both off). Using 
the CPU's test registers to clear the internal cache means that you can 
always resolve the area at fault, without needing a specially configured 
setup, without needing to touch the hardware.
 SM> I've found that for best results, I need to try several different
 SM> diagnostics programs... and even if a test fails, that apparently
 SM> doesn't always mean that there's a problem.  I've seen some memory
 SM> tests fail consistently on certain motherboards, yet the computers
 SM> work without apparent problem.
In order for a diagnostic to be useful, it must be updated regularly, as 
technology changes. If you're still trying to use a 286, 386 or 486 
diagnostic on a pentium system, you should EXPECT false results. Period.
Examples of this abound. Norton Diags v1 fails Cyrix CPU's, for example... 
why? because Ndiags doesn't recognise the cyrix CPU, thus it doesn't realise 
that Cyrix' flag regiser behaves differently, so it takes this 'different' 
behaviour to mean the cpu must be faulty. It was fixed in later versions of 
ndiags, but for *months* (Until Symantec finally fixed it) i had customers 
arguing that they were sold a 'bodgy' CPU.
My complaint is that the tools are not updated regularly, if at all. By 
writing my own, I avoid that, since I can update it as often as I need to.
The 'downside' of my tools is that I refuse to suppot anything other than 
'clean-boot' environmnt, and insist on a minimum of a '386, and since I dont 
own a Pentium-Pro or Pentium-II, 6x86, K5 or K6, I don't know if it works on 
those chips. As hardware passes thru my workshop, I get to verify these 
things, but sometimes it's a long wait!
       Craig
--- FMail/386 1.20+
---------------