The crystal time function also gets worse when voltage to
that oscillator varies. And that voltage variation is part of
the CMOS date time clock design. Oscillator accuracy, which
addressed the OP's original question, was answered at the very
beginning of this thread. Variation of oscillator crystal is
why watches have a trimmer capacitor. But this too was
answered up top, at the beginning, maybe about 100 replies
ago. Worse still, something like 3 out of 4 responses still
post about external time sources - WWV, internet time,
Operating System response to a interrupts from a completely
different clock, etc. All those were obviously and totally
irrelevant to the question originally asked by do_not_spam_me.

His question was answered early on in the very first
responses. Following posts should deal with these - two
factors of oscillator variation - no trimmer capacitor and
voltage variations to that oscillator. So many reponses that
never once even considered these factors that cause CMOS RTC
timing variations.

BTW, John Popelish also provides good information on how
crystal cuts and other factors contribute to these
variations. Another post that answers the OP's original
question and a comment about many posts that don't.

Brad Houser wrote:
OK, so we eliminated the obvious. The typical PC crystal costs $.25
and is accurate to 20ppm, or 1.7 sec/day, at room temp. It gets worse
at other temps. You can buy a pci clock card with a tempurature
compensated crystal that is accurate to plus or minus 2ppm for $60,
and make it your time server for you network, or if stand alone
systems, in each one if it is that important. Here is one:
http://www.beaglesoft.com/clcaspecs.htm