Chris,
I've got a production system running CentOS 4 that was rock solid until I upgraded from 2.6.9-55 to 2.6.9-78.0.13 (now running 2.6.9-89.0.11). The system now crashes intermittently after a few weeks. I finally caught the panic message :
EDAC MC0: INTERNAL ERROR: channel-b out of range (4 >= 4) Kernel panic - not syncing: MC0: Uncorrected Error
Looking at the kernel changelog, I see that EDAC support was added for the Intel 5000 chipset in 2.6.9-68.20.EL which this server runs.
Same issue here with a machine running centos 5.3. The problem began with a kernel update that introduced the 5000 chipset. See the thread "RAM errors after kernel-update" for more details. I couldn't solve the problem yet, but because the machine crashes every two days with this kernel, I had to boot an earlier kernel without chipset support.
I'm trying to determine if this is a potential memory issue, or is this related to some other hardware item. Also considering disabling EDAC in the kernel (is "noedac" a valid option?) as a last resort. I will run memtest86+ on the server as soon as possible to check the memory, just formulating my game plan if it's something else.
Don't use the memtest86+ version that comes with the centos ISO. There is a much newer version available from the authors website. Only the new version identifies the chipset correctly.