xoai, on 05 January 2017 - 03:22 PM, said:
Did a lot of experimentation today.
With snaps local/host and sv_fps all set to 125 the hits are amazing, but there is a crazy stutter frame lag (126 constant in gui though).
The only way to get rid of the stutter completely is everything at 20. At 20, the hits are marginal, slightly worse than in 4.2 (I realize behavior should be exactly the same in this instance, but there is an actual, tested difference)
The stutter is probably related to model prediction and its interaction with netlag compensation (inside server). It is already present in 4.2, but only lightly noticeable in some clients (64 bits ? or linux, mac) at snaps = 20.
xoai, on 05 January 2017 - 03:22 PM, said:
On popular pug servers where we get >200 (FA and East CTF namely) the hits are mind numbingly bad. Private scrim servers range from awesome to just as bad. I haven't discovered any magic numbers on our server, but 20 seems to be the most consistent across different players. Some of us got turbo hits at 30/60/120/125 and others none at all. Running snaps at half fps seems to work, but fps lower than snaps was awful.
Have you tried snaps and sv_fps=25 (with 125fps in clients)?
All time/frequency related computing are done internally by delays in integer milliseconds. 125 fps means delays of 1000/125= 8 ms. for screen updates. Also, snaps=20 (or sv_fps=20) means 20 snaps/second, or delays of 1000/20= 50 ms in network updates (server send position of other players, etc). 50 is not divisible by 8. But 25 snaps/second is equivalent to cycles of 1000/25 = 40ms and 40 is exactly divisible by 8. Maybe, having position updates in multiples of screen prediction could improve something. At least 25 snaps/second is not that much load compared to 20 snaps/second...
On the other hand, the stutter was always there but diminished by the effect of snaps capped to 20, supposing there is not a constant value of 20 snaps (or 50ms. delay) fixed in some internal computing (or implicit in some expression) instead of using the variable value setting.
The stutter (and indirectly fps drops and missed hits) seems to depend on the relation between the ping computed by server (the one shown in scoretab) for players involved. I observed (inside the same server) that most stutter (and less hits) is noticeable in other players with ping not multiple of 8ms. as I usually have ping=48 (in scoretab but 24 or 32 shown in client with cg_drawFPS).