From experience i'd say few things might be the problem
a) incompatible memory or too tight timings for dual channel to work properly (if it's dual-channel). I for example have to reduce timing on one dual-core Athlon64 from 1T to 2T or the box isn't stable even with lowered (manually) DDR-speeds. Not even with BH-5 chips.
b) CPU itself is overheating which makes gcc internal checking fail. This should manifest itself differently tho (the signal 11 syndfrome).
I do remember having those segfaults with too tight memory timings sometimes. Usually the box just hard locked every time in less than hour. The 1GB setup (2x512MB) was fine with any timing if not using dual-channel, but dual-channel just didn't work with default timings. The mobo here is DFI LanParty (something PCI-E) which has been happy with 4x512MB dual-sided DDR-400 (should be reduced to 333 tho) after diabling the 1T timings.
I can't say that this is the problem, but from the symptoms i'd pretty much say it is.
Hi Pasi!
Nice hearing you again :)
Thanks for the input. This box has been running Centos 3.5 for over a year now without any stability problems (running plain vanilla kernel 2.6). It's a single core Athlon64.
I guess that would pretty much rule out a hardware problem? I'll try to reduce the timings anyway.
Cheers! Remco