can anyone confirm? computer resets when loading winx, not softw, one bad address when testing good ram on this machine

Apparently bad MB but has anybody ever seen this? I have tested all components in the system
to be good by swapping with other machines and diag softw. (except processor, MB, ps).
Tried clean install of xp and me - boots fine off of install cd, but computer resets when trying to
load win after file installation pass. Selfbooting Memtest2.5 and memtest86+ show one bad
memory address, (ram tests ok on other machines), selfbooting ibm diagnostic does not find
anything wrong. Booting from ERD 2003 cd does same - after loading files resets computer
instead of entering 32bit mode.
I flashed latest bios. I am an experience tech but have never seen this.
Is it possible for the MB or CPU to not be able to address one bit of good RAM out of entire
range?

Its an IBM Netvista Celeron 900, no OS, brought to my shop for eval/possible repair. There is
no known history with it. All components are eliminated as possible causes except - MB, CPU,
PSU. Its not the RAM as I have numerous known good chips to try. All test good in other
machines and pass stand alone software RAM testers that actually work (www.memtest86.com,
www.memtest.org, hcidesign.com/memtest - all are free, boot themselves from floppy and are
independent of your OS.) All these memory modules fail at the same address only when testing
with one of these three programs.

Same problem at same point when attempting to installing XP, ME and 98SE).

I can boot to DOS and to the self booting diagnostics. As I mentioned, cannot boot using
ERD 2003 which is a high end diagnostic tool which loads the basics of an XP/2003 install
entirely from its CD ROM. It resets after loading itself into RAM and at the point it attempts to
enter 32bit.

I can't isolate the CPU from the MB because I don't have any other .18 micron socket423
(370PGA) celeron processors or another MB of that vintage on hand.

The reset occurs regardless of SIMM socket used.

The simple answer is that there is an addressing problem in the MB that occurs when the
address range used to enter enhanced mode is used for the first time. The self booting IBM
support diagnostics do not confirm this however.

I am just asking if anyone has ever seen this before. The customer wants to part system out
rather than purchase MB and/or CPU.

Its just bugging me. Like I'm missing something. And yes I do have a habit of working way too
long on periodic mysteries.

Thank you for all responses.