[Ti] any comments on memtest v2.95, a command line memory testing utility

[Kynan Shook]

I'd say it looks worse than many software tests, simply because it can 
only run while the OS is running, requiring at least 100 MB of RAM to 
never be tested, by the porter's own admission.  If it's written well, 
then you can believe any failures it may indicate.  But it will still 
probably only catch about 5% or less of bad RAM.  Also, I don't like 
the idea of writing random values to RAM as one of its tests; unless it 
always uses the same seed value, this means the test is not repeatable, 
which is A Bad Thing.  If it does use the seed value, well, then it's 
not writing random values to memory, now is it?
I stand by my statement that there is no (and, for that matter, never 
will be) a good software memory test.  The inherent problems are that 
the operation of the test requires some of the memory to run itself, 
and that many aspects of the test are not controlled by the test 
program: A hardware RAM tester has its own memory separate from the RAM 
being tested, so every bit can be tested.  Also, a hardware RAM tester 
can do things like fluctuate the supply voltage, briefly remove power 
from the RAM, change the frequency it is running at, run extra current 
through the memory to heat it up, and play around with other things 
that are normally very tightly controlled (and not able to be changed) 
in a computer.

The moral of the story is that most memory tests will be correct when 
they say memory is bad (there are some exceptions; for example, memtest 
might tell you your memory is bad if, for example, your CPU is actually 
the part that is fried; the best way to tell would be to see if you 
keep having the same memory error even with high-quality memory 
installed).  However, software tests probably won't catch more than 
about 5% of bad RAM due to the limitations of the environment.  The 
hardware tester that I use probably catches around 75% or so, at a 
price of around $5,000.  If you want to get closer to 100%, you can 
spend $500,000 and get a tester with a variable temperature oven, 
accuracy down to fractions of a nanosecond, and all that...  If you 
really want to test memory even better, you have to know the internal 
physical layout of the bits in the module, which is something only the 
chip manufacturer knows.  But even all those improvements can't account 
for all possible ways for memory to fail, just the most common ones.  
The only true test of memory is to see if the computer crashes less 
often, has fewer directory corruption problems, etc. if the memory is 
removed.

bofus? <bofus at mindspring.com> writes:
> 	are there any thoughts about how good this ram tester program is?
> 	http://www.versiontracker.com/dyn/moreinfo/macosx/17156
> 	"memtest v2.95, a command line memory testing utility"
> 	curious...