AVR-Chat

>  David has described how to pick the load capacitors if a) the crystal
>  is parallel reasonant (otherwise, the data sheet won't give a
>  capacitance), b) 5 pF is a good estimate of the parasitic capacitance
>  at each terminal (I don't think it is, according to the ATmega32 data
>  sheet, the pins themselves can have up to 10 pF of capacitance)

The chip capacitance, like the loading caps, are in series, from the
crystal's point of view, so 5pF is reasonable.

And
>  c) you really, really care about the exact frequency to withing a
>  hundred parts per million.


The problem with more loose methods is that the crystal may not start,
or it may run on other modes.  You really should do a full design,
including measuring the drive level and oscillation margin.  But few
do.


>  A more pragmatic way to pick the load capacitors would suggest that
>  22 pF is just fine.  To generate the bit rate for a UART, you only
>  need to be close, not dead-on.  Of course, 27 pF is just fine too.
>  Which gets us back to the point that the frequency should not swing
>  wildly due to this change in the load capacitors.

Definitely not.