For interactive systems, such as computer workstations and video games, the threshold of acceptable input-to-output latency is dependent on the type of physical input.
Furthermore, video latency is only one portion of the total input-to-output latency. The source device‘s processing and transport make up the remaining latency, but often can not be controlled by the consumer, e.g. in the case of console video games where only the sink devices can be chosen and configured.
Type of Physical Input
For a user to feel that their physical input to an interactive system results in an instantaneous response, the total input-to-output latency must be below a certain threshold. This threshold is entirely dependant on the type of physical input being performed.
For example, the response to a button press is said to feel instantaneous if it becomes visible less than 100 ms after the button press.1 Conversely, the response to finger movement on a touch screen can be noticed above 1 ms.2 Mouse movements lie somewhere in between.3 Video game “motion controls” likely match a similar threshold to mouse movements. Virtual reality requires less than 20 ms of latency between head movement and resulting image.4 These thresholds also vary between different people.
Source Device’s Processing and Transport
There are many elements that contribute to the entire intput-to-output latency of an interactive system. Each of these elements often have a variable latency that is dependent on the exact timing of the user’s input.
Example: A button press in a theoretical 60 Hz console video game:5
- Controller signal processing (debouncing, etc.): 5 ms
- Transmission and kernel level processing: 4 ms
- Waiting for next game state processing (variable from 0 to 16 ms): 8 ms
- Game state processing: 16 ms
- Game rendering: 16 ms
- Time to transport half of a video frame: 8 ms
- Video latency: 0 ms
Total input-to-output latency: 5 + 4 + 8 + 16 + 16 + 8 + 0 = 57 ms
While the following theoretical example is important to gain an understanding of the complexities at play, it is even better to look at some real world examples of interactive systems, such as console video games. Here is a video that reports the minimum total intput-to-output latency when using a display that has only 1 ms of video latency:
As you can see in the example video, it is not unheard of for some console action-genre video games to have more than 95 ms of input-to-output latency.
It is important to note that the measurements reported in the video are the minimum, so it can be expected that maximum latency measurements would be around 16 ms higher in these 60 Hz video games. These measurements were performed at the top of the screen, so 8 ms can be added to account for the time to transport the first half of the video frame. Another interesting point is that intput-to-output latency can become higher in newer generations of video game consoles.
What is an Acceptable Sink Device Video Latency?
To account for all different types of physical input and different source device processing and transport times, it becomes clear that even a display with 0 ms of video latency may not enable user input to feel instantaneous. When choosing a display that may be used for a wide variety of interactive systems, e.g. different video games and consoles, just a few milliseconds of video latency could be all that is needed to cause a visual response to no longer feel instantaneous. For this reason, a display with as close to 0 ms of video latency is always recommended.
Author’s note: Given technology that is available in 2021, my personal recommendation for a display that will be used in an interactive system is one with 4 ms or less of video latency for enthusiasts and 8 ms or less for others.
Units of Measurement
A millisecond is a very small period of time, so small that is imperceptible in most situations. For this reason, and because accurately measuring latency can be very difficult, it is standard in the engineering and science fields to report latency values for interactive systems and audio/video synchronization as whole milliseconds.
Last updated on February 2nd, 2022.
- Response Time in Man-Computer Conversational Transactions, Robert B Miller 1968 “Topic 1. Response to control activation”
- Microsoft Research Applied Sciences Group: High Performance Touch
- Are 100 ms Fast Enough? Characterizing Latency Perception Thresholds in Mouse-Based Interaction
- Qualcomm OnQ Blog “Keeping the virtual world stable in VR”
- These numbers are rough approximates that should match somewhat closely to a real-world example.