Mem error follows address, not DIMM ; how to cope?

Question

I have occasional BSOD that concerned me due to increading frequency. Once in two years, then 4 months, then 2 weeks.

I ran Memtest86+ and it locked up after 6 minutes 20 seconds. It repeatedly locks up in the same place (not universally; outlier was earlier). I tried different botherboard settings including Safe Defaults. Having reproduced the issue with a baseline setting, I tried reversing the two banks, each bank installed alone, and always got the same address, in the 4 to 6 MB range if I remember correctly. It follows the address, not the physical RAM module.

I tried one bank only populating the second, with no difference.

Memtest86+ does work OK in F1 Failsafe mode, it seems (longer test run happening now). Note that turning off a couple cores did not make a difference, but I'm not sure if the indicators mean that just one or many cores were engaged; the display might not ahow the hit that caused the freeze, anyway.

Any idea what might be causing this? And more important, how might I mitigate it until I get new hardware set up?

Can I get Windows to block out that range and not use it? (edit: yes! See https://superuser.com/a/490522/368845 )

Hmm, would PCI cards be doing something in that range, or does memtest86+ only count normal RAM?

---

The CPU (with integrated memory controller) in the AMD 1100T.