[Techtalk] Excessive clock drift

[H. Peter Anvin]

Stephanie Boyd wrote:
> Hi,
> 
> We've just got a new Athlon 1600+ (1.4GHz and 133 FSB) and a GA-7VKML
> motherboard and I'm seeing excessive drift in the system clock (just
> over +1 second per minute) which is greater than ntp can correct.
>  
> The problem seems to be with the system clock, as the following output
> shows the difference between the hardware clock and the system clock
> is close to that reported by ntp.
> 
> root at ixchel:jjm# date; hwclock --show; ntpdate -q ntp.demon.co.uk
> Wed Oct 16 22:26:23 BST 2002
> Wed Oct 16 22:26:19 2002  -0.478815 seconds
> server 158.152.1.76, stratum 2, offset -5.061046, delay 0.05101
> 16 Oct 22:26:24 ntpdate[3408]: step time server 158.152.1.76 offset -5.061046 sec
> 
> I should mention that this has been seen under kernels 2.4.18 and 2.4.19
> (which we've been running successfully on other machines for some time).
> 
> Has anyone seen this type of clock drift on Athlons before ? 
> 

I have seen it on various systems; basically you have a good clock 
advancing at the wrong rate.  You can use the adjtimex utility to 
configure the kernel to expect the particular rate your system clock is 
actually advancing instead of the default, nominal frequency (you have 
to set up adjtimex to run on every boot -- it's not a permanent change.)

Once you have set the rate with adjtimex ntp should be able to catch up.

This is pretty common; the master crystal in PCs is virtually never 
calibrated, and sometimes can be off by as much as 2%.

	-hpa