> What kind of accuracy can one expect from a PIC with a X MHz crystal?
> If you assume that the crystal is off by 25ppm, you would be off by
> 0.000025 seconds per second.  This would mean gaining (or losing)
> 0.000025*3600*24=2.16 seconds in a 24 hour period.  This does not
> seem very accurate.
>
> >
> > [2] Use any PIC with a 32768Hz clock.  This will drop power consumption
>
> Is the 32768KHz crystals used for RTC's more accurate than the 'normal'
> crystals?  Or do you get the same kind of accuracy from them?

There really isn't anything magic about the oscillators on RTC chips.  There
are various factors that will make crystal oscillators of any type run faster
or slower than intended.  Some will be fairly constant for given devices and
component values; some will vary from one device to another but be relatively
constant under temperature changes; some will vary with temperature (but may
be consistent from one device to another).

Depending upon your accuracy requirements, you may opt to:

[1] Ignore the problem--let the crystal run as it will

[2] Tweak the capacitors so that the "average" unit runs well

[3] Tweak the capacitors on individual units so they all run well (many RTC
    chips let you program a "trim" value which varies the load capacitance
    slightly)

[4] Adjust in software (either an "average" amount for all units, or on a
    measured-per-unit basis) for the clock being fast or slow.

[5] Adjust in software, using a temperature sensor and measurements made on
    each unit.  This last approach requires a fair bit of work (since you
    have to measure each unit's precise speed at different temperatures) but
    it is *extremely* precise.

The first method is the easiest; the last is more precise.  Use whatever seems
appropriate for your application.