[CentOS] Re: Memory problems with CentOS box

Sat Oct 20 10:55:59 UTC 2007
Kay Diederichs <kay.diederichs at uni-konstanz.de>

Miguel Medalha wrote:
> Hello all
>  
> I am running CentOS 5 on a small server and I am having very strange 
> memory malfunctions.
>  
> The computer runs perfectly with no problems whatsoever. From time to 
> time, after a soft reboot, the computer emmits beeps corresponding to a 
> memory fault.
> It never reboots again until I find and remove a now defective DIMM. 
> That DIMM can never be used again because it is out of order.
>  
> This just happened for the *fourth* time and is costing me a lot of 
> hassle and perhaps expense if the manufacturer does not replace the DIMM 
> freely.
>  
> The memory is DDR400 ECC from Kingston and of the type recommended by 
> the motherboard's manufacturer. The board is a Tyan Tomcat i875p (S5102).
>  
> Would it be possible that the DIMMs are being destroyed by some software 
> component from the OS, perhaps the I2C management? The DIMMs do have 
> EPROMS... Are they being incorrectly accessed by some software component 
> and their program modified?
>  
> Could this be the board's fault? And how?
>  
> I don't know what to think of this, I never saw anything like that in my 
> already long experience with computers...
>  
> I must say again that this memory error *never* happens while the 
> computer is in service, it always happened upon a soft reboot.
>  
> Any hints would be appreciated.
> Thank you
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> CentOS mailing list
> CentOS at centos.org
> http://lists.centos.org/mailman/listinfo/centos

I've had similar problems with a Tyan motherboard (8-Opteron S4881/S4882 
combination) which had all memory slots filled.
It turned out that the BIOS allowed to run the memory at its full DDR400 
speed, whereas the AMD specs say that one has to go down to DDR333 
speed. The result was destroyed DIMMs _and_ Opterons (the built-in 
memory controllers)!
It was not easy to track this down; I'm still wondering why the BIOS 
programmers do not read and follow the specs!
This computer is now happily running at DDR333 with uptime > 6 months.

The upshot: don't trust Tyan BIOS - check possible settings against 
specs. Don't trust vendors - they sell DDR400 memory even if the 
particular setup only allows to use DDR333 speed.

HTH,
Kay