This is a screenshot of memtest86+ running only test #5 (Block Move) on the module MT19721975197704. Execution time was
just over 11 minutes ("WallTime") and in that time, four complete passes were performed ("Pass", middle right). There
were errors detected during passes 2 and 3 ("Pass", far left column). The failure was limited to two different addresses
("Failing Address") and in all cases the same output pin was bad ("Err-Bits"; failing pin indicated by "1").
The same module running the "standard" tests. This is a group of about 10 different "patterns" that are executed on the
memory module. For a 512MB PC133 module, it takes about 45 minutes for one complete pass. In this picture, the module
had been tested for 2 hours and 14 minutes. In that time, two complete passes were finished and it was starting a third.
The only two errors were again from Test #5 and were at the identical addresses in the first picture, and were failing
for the same output pin.
It is my belief that there is a marginal memory cell in this module which can fail during some very specific data sequences in the 64 bits. This failure would be random in nature and will probably result in a "blue screen" due to the corrupted data. I ran the same module, in the same T23, using the diagnostic program supplied by IBM ("PC Doctor") and it passed those tests (about 30 minute run-time).