"Walter Clayton" wrote:

BTW: Using software based memory testing doesn't work very well within a
running OS.

Oh yes? The Memtest program I mentioned sure found some problems with
my memory (Corsair PC3200 memory rated at 2-2-2-5). It turns out that
to use such timings, one should increase the memory voltage to about
2.75 volts. There are special "heat spreaders" (Corsair's term, not
mine) on the RAM modules to dissipate the extra heat. Unfortunately,
there is no such voltage information on the memory package. Only in
digging through the Corsair web site after Memtest showed the problems
did I find the data sheet for the specific ram. The data sheet
indicated that "all tests were performed at 2.75 V". I later learned
that the need to increase the voltage a bit for such "special" low
latency ram is common knowledge among "overclockers". Once I upped
the voltage to 2.7 V, the memory errors went away. As a margin test,
I then temporally overclocked the machine (CPU + FSB) +5% and tested
with Memtest again. With no errors detected, I lowered the clock back
to normal where I'll keep it.

So, you see, the Memtest program worked quite well for me because it
revealed a memory problem (averaging about 1 error per 100 MB tested)
that was corrected when I upped the memory voltage to the
manufacturer's specified voltage. It should be noted that it is just
such occasional RAM errors that are the most insidious.