Low input latency is key to a satisfying experience in competitive games, but reasoning about latency can be non-intuitive in modern engines with multiple parallel timelines running at once. This talk will tackle latency by analyzing the path that data flows through a game engine from the input sample all the way to the video scan-out hardware. By identifying and measuring segments on this path, a dynamic throttle can be implemented to squeeze the input sample as close to the end of the frame as possible without dropping frames. Special attention will be placed on how latency is introduced in the interaction between the GPU and display scan-out timelines when Vsync is enabled.

Takeaway

Attendees will gain insight into how input latency works in games engines, especially how it can be introduced between the GPU and video scan-out. They will learn a strategy for implementing a throttle on the input sample to reduce latency, as well as the trade-offs to consider when tuning the throttle.

Intended Audience

The talk is intended for programmers interested in game engines, and most of the concepts should be applicable to all game engines large and small. The content is skewed towards a rendering perspective, but no specific rendering knowledge is required.