Tickrate Scaling and Physics Simulation Step Decoupling: Historical Paradigm Evolution and Next-Gen Architectural Projections

[Elijah Campbell]

So I’ve been digging into how we’ve moved from fixed-timestep physics locked to the tickrate toward decoupled simulation steps, and I want to hear from people who’ve actually shipped or tweaked netcode, not just theorize. When you’re running a 128-tick server, do you still pin the physics step to the tick interval, or have you moved to a fixed accumulator with a max substep cap to avoid the spiral of death? And more importantly — where are you drawing the line between simulation and interpolation now that you have multi-threaded networking? I’ve seen teams keep the physics on a dedicated thread separate from the networking socket thread, but I keep hitting input latency penalties when the sim thread gets starved.

Here’s what’s really bugging me: with variable render interpolation now basically standard, how are people handling the prediction reconciliation when the physics and render loops are completely decoupled? Are you seeing meaningful differences in the hit registration feel between a system that rolls back the world state versus one that just lerps entities? And for anyone pushing into next-gen — is anyone actually running fixed-timestep at 64Hz with higher-frequency interpolation, or is the community still chained to the old 64/128-tick dogma? Curious what’s actually working in production versus what’s just theory.

"Precision over speed. Consistency over luck. Stay disciplined."

[Author]

Posts