[CentOS] Re: segfaults with 8 gig of ram

Scott Silva ssilva at sgvwater.com
Sat Mar 24 18:25:10 UTC 2007

tblader spake the following on 3/24/2007 6:46 AM:
> Feizhou wrote:
> <snip>
>>> Right now, it's been running for about 40 minutes (the Java install
>>> loop mentioned previously) with the dram clocked down to ddr-533.
>>> All four sticks are in and I ssh'd in and started another java install
>>> loop over the wire just to aggravate things.  If this continues to
>>> run for a few hours I'll call it good enough.
>> So all four sticks, clocked down and kernel parameters = A-OKAY
>> (assuming it is stable)
> Almost. Much to my dismay, the 4 stick testing went well but the
> BIOS screens occasionally hung during reboot through the course
> of the day.  I finally just pulled the 4th stick and went home.
> I've modified the rc3.d/S99local* script to just reboot every 10
> minutes and compile a kernel and modules in between.  I thought
> I'd heat things up a bit piping urandom into /dev/null as well.
> I've got other fx-62 boxes running the same mainboard with all 4
> ram slots full, but those are Corsair ram @1G each - the ram timings
> are different as well; 5-5-5-12 for the Corsair and 6-6-6-18
> for the GSkill.  Maybe the GSkill ram just isn't up to the task
> running Dual channel.  I have the weekend to play with it so maybe
> I'll try two sticks in Dual Channel for a while and see what happens.
> I'll post kernel compile times between the 3 sticks in single channel
> and 2 in dual channel :)
I bet the bios of the board is either not honoring the SPD settings of the
memory, or the memory just isn't a good match for the board. I have had this
problem before, and now I usually only buy memory that is on the recommended
list of the manufacturer, especially on critical systems. It just isn't worth
the extra trouble to maybe save a few bucks. Your time tracking down the
problem probably has already gone over the difference in memory cost.


MailScanner is like deodorant...
You hope everybody uses it, and
you notice quickly if they don't!!!!

More information about the CentOS mailing list