HyperVision / Interactive timing lab

From key press
to video frame.

Follow one input through the physics engine and onto the screen. Compare the original 20 FPS renderer with HyperVision’s 60 FPS interpolation.

Both sample gameplay input and advance physics at 20 Hz: once every 50 ms. HyperVision draws three visual steps between completed states. Smoother motion does not mean faster input sampling.

02 / Read the timing

First response ≠ full new state.

Input waits for physics.

A held input arriving between ticks waits for the next 50 ms tick in either renderer. The ideal sampling wait is from 0 to just under 50 ms, averaging 25 ms for uniformly timed presses. The extra HyperVision frames do not run extra authoritative physics or record extra inputs.

Video takes different steps.

The original renderer shows the completed state immediately, then holds it. HyperVision first shows a ⅓ blend from the previous state toward the new one, then ⅔ at +16.7 ms, then the full state at +33.3 ms. Those fractions describe the visual blend, not the strength of your input.

Smoothness has a time offset.

With no extra delays, the first changed frame arrives at the same time in this model. The full new state takes 33.3 ms longer with HyperVision. At each 60 FPS presentation, its visual sample is about 33.3 ms behind the physics timeline; the original instead holds each frame for 50 ms.

Model assumptions and source code

This is normal driving at 20 Hz with established interpolation history and sustained target frame rates. Physics calculation and rendering are treated as instantaneous so the scheduling delays are visible. The optional display delay is a fixed offset after each frame is prepared.

Real input delivery, computation, rendering, VSync, scanout, and the monitor add delay. Slow frames can also postpone input processing or skip visual samples. The values here are not measured key-to-screen latency.

HyperVision interpolates completed states; it does not predict future physics. Startup, seeking, pausing, rewinding, and camera changes can reset its history. Confirmed crash transitions can show the current pose immediately. The separate 10 Hz simulation mode has about 83.3 ms of interpolation delay and is outside this 20 FPS comparison. Desktop double buffering does not itself queue an extra game frame.

The model follows the project’s timing documentation, race presentation loop, and input sampling code.