Re: My Computer Drifts!


On 20220604 02:25:26, David J Taylor via wrote:
On 04/06/2022 08:31, Dirk wrote:
David, talking about GPS:
With NTP alone you can reach a time deviation in the lower millisecond range.
Using only the GPS time information you CAN reach a time deviation also in the
low double-digit millisecond range, if GPS hardware and software can be told to
decode only ONE NMEA sentence.
With GPS alone or with NTP plus GPS you CAN reach a time deviation in the
microsecond range, using certain GPS hardware.
If you e.g. read the Meinberg documentation, the use of the GPS *PPS signal* is
described in detail. With the PPS signal the beginning of each second can be


I think may Linux users would be disappointed to get low milliseconds from NTP
- and Windows makes it much more difficult (although better with Windows 8 and
later).  The problem with GPS alone is that while you may get a low deviation,
that deviation is referenced to other than UTC - certainly having a single NMEA
sentence and high baud rate can help.  I've seen differences in excess of a second!

GPS alone is not good enough for microsecond deviation, you must have PPS as
well.  A GPS over USB, for example, does not typically have a PPS signal, and
even if it does it subject to the USB polling rate (125 microseconds).

Getting good results can depend on pot luck, to a degree, as different network
cards and drivers influence the results.  Just for interest:

Windows stratum-1 PPS-synced PCs:

Windows (& Linux) synced over the LAN from stratum-1 servers

SatSignal Software - Quality software for you
Email: david-taylor@...
Twitter: @gm8arv

My configuration here involves a Linux router that does a lot of other work and a largish set of PCs around the house, mostly here in the computer room. They range from PLCs to a rather hypertrophied gaming machine use for multiple VMs, development, and the like. (Visual Studio is THE preeminent game in the known universe.)

When speaking of accuracy you have to deal with a pot load of different entry points for errors. The ones NTP monitors and considers in its design are delay, offset, and jitter. NTP, as embodied in the canonical Univ of U of Delaware version as amended up to date, measures the path delay in such a way as to get a reasonable estimate of path delays to multiple sources. The path to and the path from the server may be slightly different. You can see that in the offset value. NTP makes a best estimate for time and publishes the time offsets. It also passes this information along to the consumers. Offset here varies from -5.3ms to 5.9ms. The chosen clock and the secondary are coming in at + and - 2 ms. The chosen clock is .NIST.. The secondary clock is .GPS. I could setup a pure GPS configuration here if I hooked up the PPS from my Trimble to the router. But, it's not broke so I don't fix it. The final figure, jitter, gives a good estimate of the link quality between the consumer and the hosts. The two good ones are coming in at under 6 ms of jitter. All these numbers vary with time of day. For my current needs this is good enough. If I needed better I'd have to go to a formal PPS hookup. (None of the USB crap, either. They do look like fun to play with, though.)

For my main PC I am using a very old tool I have had for a long time. Mr. Horsley created an open sores partial implementation of ntp that I use to tweak my PC's clock in 1ms steps - since at least early XP days through today. I have it logging clock corrections. Most are 0 ms with the occasional 1ms correction. I ran Symmetricom's ntp for awhile. But I went back to Tom's tool. It ain't broke and I have a much older view on the urgency to make perfect than I used to have.

(I am one of the persons responsible for the GPS phase 2B satellite segment's extremely accurate intentional inaccuracy for whatever worth that may be.)

If ntp accuracy is insufficient at least one more protocol exists that refines to microseconds or finer. It is/was used by radio astronomers to get them in the ballpark.


Join to automatically receive all group messages.