Summary
Under high latency (~100+ ms), Server Authority input responsiveness degrades sharply as player count increases. With 1β2 players the experience is acceptable; past that, input delay grows with each additional player and becomes effectively unplayable around 4+ players. The scaling with player count, rather than with latency alone, suggests client-side prediction is failing or being discarded rather than this being a pure latency issue. Excessive rollbacks are also present, and are especially pronounced when a new player joins.
Environment
- Client ping: ~100+ ms (verified by connecting from the US to a Brazil server)
- Player count: 3+ players in the server, worsening with each additional player
- Reproducible on a blank baseplate with default character/SA settings
- Not yet verified in Studio β Server Authority behaves noticeably differently in Studio than in the Roblox client, so all testing was done in a live client session
Steps to reproduce
- Publish a blank baseplate with Server Authority enabled.
- Join from a client with ~100+ ms ping (e.g. connect from the US to a Brazil server).
- With only 1β2 players in the server, move around and note baseline responsiveness.
- Add additional players. Observe input latency increasing with each player added.
- At ~4 players, movement input becomes severely delayed to the point of being unplayable for a competitive game.
- Have a new player join mid-session and observe a burst of rollbacks/corrections for existing clients.
Impact
Any experience (especially competitive) using Server Authority becomes unplayable for high-ping players once the server has more than a few players β which is the normal case for most games.
Expected behavior
Client-side prediction should keep local input responsive regardless of how many other players are in the server. Input latency should be a function of local prediction quality, not remote player count, and other players joining should not trigger rollbacks of the local playerβs own predicted state.
Actual behavior
Input delay scales directly with player count at high ping, reaching unplayable levels around 4 players. Rollbacks occur far more often than expected, with a distinct spike whenever a new player joins the server. The pattern is consistent with prediction either not running or having its results repeatedly invalidated as the number of simulated characters grows.