[LinuxPPS] Strange offset behavior

[Hal V. Engel]

On Tuesday 16 June 2009 01:47:48 pm Andrew Hills wrote:
> > The oscillators on these machines are free running quartz crystals with
> > no temperature compensation or control. ...
>
> Thanks for all the information. In your personal opinion, if three of
> four machines exhibit the more consistent behavior and the hardware in
> the fourth differs from the other three only by the motherboard
> (completely different manufacturer/model), is it safe to assume that the
> wild offset behavior is due to poorer quality hardware in that model?
>
> --Andrew Hills

Andrew,

It is hard to say.  But one of the things I have noticed looking at 
motherboards is that they vary significantly in the location of the oscillator 
crystal.  Some of these have this located very near heat sources (CPU, 
northbridge, southbridge...) on the motherboard and others have these located 
well away from these heat sources.  This could (likely does) have an affect on 
how stable the temperature of the crystal is and this could affect time 
keeping.  But I am not sure which design is better for time keeping.

There are a lot of variables here and I suspect that motherboard 
manufactures/designers have significant constraints in this area and that this 
is affected by things like the CPU and motherboard chip sets being used.

The motherboard designer does not have time keeping as a major design goal.  
In fact it may not even be on their list at all.  I don't think any 
motherboard manufacture screens their crystals in any way to assure that they 
behave in a way that is consistent with time keeping goals.  I also think that 
these crystals are specified in a way to keep costs low and this likely makes 
them poor time keepers.  It is well known that the crystals used in quartz 
watches, even really cheap ones, are more stable then those typically found in 
computers by a significant margin.   The reason for this is that the crystals 
in watches are manufactured for time keeping and are build and tested to a 
much stricter specification that those used in computers.  They are also more 
costly.

Since time keeping was not one of the goals of the motherboard designers I 
don't think you can make any statements about the quality of the hardware 
based on the stability of the oscillator.  In fact about the most you can say 
is that the oscillator in machine X in not as stable as those in machines W, Y 
and Z and therefore machine X might not be a good machine for time keeping.    

Hal