NVPPS PTP synchronization

I guess just to close this out:

  • Does NVIDIA see this as a problem?
  • If yes, does NVIDIA plan to do anything about it?

I realize that 0.6 microseconds per second is a pretty slow drift in the big scheme of things, but it makes a lot of this new fancy cam_fsync library functionality considerably less useful, particularly as NVIDIA has spent time and energy getting TSC to syntonize to PTP via the nvpps driver.

(Also note, in the documentation surrounding TSC, I would recommend using the term ‘syntonize’ instead of ‘synchronize’, as TSC does not synchronize to PTP but does adjust its definition of a second to match the PTP second - that’s syntonization)

This problem can be alleviated by using slightly modified configuration values.
For the 30Hz case in Orin, the problem can be alleviated by defining a pattern that covers 3 periods of the 30 Hz, with 2 periods at 1,041,667 x 32 ns and 1 period at 1,041,666 x 32 ns. The total is 3,125,000 x 32 = 100,000,000 ns exactly 10 Hz, so the pattern does not drift against the reference.
This makes one pulse slightly off (by 1 TSC unit) but this should generally not be an issue.

However, the current driver, cam_fsync.c, does not implement it yet.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.