[time-nuts] Re: World's most accurate PC clock!
Tom Van Baak
tvb at leapsecond.com
Sun Jul 3 14:12:24 EDT 2005
> A maser-driven synthesizer is definitely the best approach, but the
> frequency isn't exactly 14.31818. It's 14.3 (18)repeated, as the
> frequency is derived from 5.00000MHz * 63 / 22. That means you need
> a divide-by-eleven to get it spot on. This is all described in the
> Wikipedia entry for NTSC.
I know what you're getting at. At one level you're
correct. But since we're time-nuts allow me to
digress into precision and history here. You see,
spot on isn't good enough in this case. For a PC
most NTSC articles are close but not the whole
The frequency of a perfect "3.58 MHz" colorburst
crystal is 3 579 545.454 545... Hz (as a decimal
approximation), or 5 MHz * 63 / 88 (as an exact
rational number). Now divide this by 3 to get the
1.19+ MHz i8253 timer frequency. And divide this
by 65536 to get the 18.2+ Hz DOS timer interrupt
frequency. Since we're talking masers, the actual
value, to 14 digits, is 18.206 509 676 846... Hz.
So far so good.
But the other thing you need to consider is that
DOS keeps time and date by counting an integer
number of these timer interrupts per day. That
integer is 1573040.
As you can probably guess, the colorburst-derived
18.206... Hz frequency doesn't divide evenly into
86400 seconds per day, so the DOS calculation
of hours, minutes, seconds, and days is not perfect.
There's a small fractional interrupt residual that is
lost and ignored.
The actual number of 18.206... Hz interrupts in a
86400 second day is 1573042.43607954545...
This means that even if you were to use a perfectly
accurate atomic clock for the PC xtal reference DOS
would still appear to be fast by the equivalent of
2.436... interrupts per day, which comes to about
133 ms / day, about 1 second a week, an error of
about 1.5 ppm.
This is not a problem in practice since no DOS PC
has a crystal accurate and stable enough for anyone
to notice. But if you were to replace the PC xtal
with an atomic oscillator you'd have to take this
correction factor into account.
I also suspect the correction, if any, would be
different for other OS's, depending on how well
the OS clock software models the unavoidable
accumulated fractional error.
> Just think how easy it would be to sync up all your computers if
> their clocks were all driven from one stable source!
There are more worries besides stability. Many OS's
have a problem, under conditions of heavy load, of
missing a clock interrupt every now and then. When
this occurs it spoils the benefit of having a ultra-high
More information about the time-nuts