kernel: Uhhuh. NMI received for unknown reason...

[Alan Cox]

> Hence logic says that something that is being change between kernels
> are causing this otherwise the kernel would ALWAYS being reporting
> this...

Unlikely - hardware problems are often dependant on alignment of objects
and other chance happenings. If you've got bad RAM and the faulty bits happen
to land in a location where the faulty bits don't show a fault (its often
combination based) you'll see exactly what is described.

> If anyone can tell me how I can *debug* it further
> I'm all ears..

memtest86 full night run is what I usually start with for such cases.