Entity Interest Management is where you win or lose your bandwidth budget, so nail the spatial partitioning before obsessing over delta math. I use a cell-based grid keyed to each client’s render distance, culling anything beyond a 32-64 meter radius on a standard competitive map. The key is decoupling the update-grid resolution from tick rate. On a 128-tick server you push 128 snapshots a second, and naive per-entity delta blows past a 25-50 KB/s per-client budget fast. Capping the interest query to a fixed timestep—refresh the mask every 3-4 ticks instead of every frame—keeps snapshot sizes predictable without visible pop-in.
Bit-packing your change flags is the highest-leverage delta optimization. Instead of sending a full 32-bit flag per attribute, pack related booleans (visible, position changed, velocity changed, health changed) into a single byte mask and only serialize fields whose bit is set. That typically cuts per-client entity traffic 3-4x versus naive delta. Track a reliable last-good tick per entity so clients resume from the last confirmed state instead of demanding a full snapshot. For receive-side interpolation, hold a fixed 100-150ms buffer—less and you rubber-band under packet loss, more and effective input latency creeps past the 60ms sweet spot.
Profile with a real bandwidth monitor and target stable snapshot variance. A peak-to-median ratio above 2:1 means your interest boundaries are leaking entities into clients that don’t need them.
🎮 Battlestation & Rig: Ryzen 7 9800X3D | RX 7900 XTX Taichi | 32GB DDR5 6000 CL28 | Corsair 5000D | BenQ ZOWIE XL2566K 24.5″ 360Hz DyAc+ | Moza R9 V2 Direct Drive + Moza FSR Formula Wheel + CRP Load Cell Pedals | Audio-Technica ATH-R70x + Schiit Magni/Modi Stack
📍 Birmingham (UK)
"Under 55 grams on glass skates: zero friction, pure aim."
