Everything in gaming hardware is marketed on frames per second. It is the number on every benchmark chart and it is an incomplete description of what determines whether a game feels responsive.
Average frame rate hides the problem
The first issue with the headline figure.
An average of a hundred and twenty frames per second can contain regular drops to thirty, and those drops are what you perceive. The average conceals them entirely.
Which is why benchmarks increasingly report percentile figures — the frame rate at the first percentile and the tenth percentile, describing the worst moments rather than the typical one.
Those numbers are far more predictive of whether a game feels smooth. A configuration with a lower average and a higher one percent low will feel better than the reverse.
Frame time consistency
The related and more precise way of looking at it.
Frame time is how long each frame takes, measured in milliseconds. Steady frame times feel smooth. Variable frame times feel like stutter even when the average is high.
A frame time graph tells you more in one glance than any average, because the spikes are visible.
Stutter from inconsistent frame delivery is perceptible at levels that barely move the average, which is why two systems benchmarking identically can feel very different.
Latency is the number that actually matters
The measurement that determines responsiveness and that is almost never quoted.
The relevant figure is total system latency — from a physical input to the corresponding change appearing on screen.
That chain includes the input device polling rate, the operating system, the game engine, the rendering pipeline, any buffering, the display's processing, and the display's pixel response.
Frame rate affects one part of that chain. Several other parts can dominate it.
Which is why a system running at a high frame rate through a display with heavy processing can feel less responsive than a lower frame rate on a display with low latency.
The settings that affect latency most
Practical and mostly free.
Display mode. Televisions in particular apply substantial processing that adds latency. A game mode setting bypasses most of it and is frequently off by default, and the difference is large.
Buffering settings in the game. Queuing more frames in advance smooths delivery and adds delay. Reducing the queue reduces latency at some cost in smoothness.
Synchronisation. Traditional vertical sync eliminates tearing and adds latency. Variable refresh rate technologies achieve much of the same result without the penalty, and are worth enabling where both the display and the hardware support them.
And frame generation techniques, which increase the displayed frame rate by inserting generated frames. These improve visual smoothness and do not reduce latency, and in some implementations increase it, which is a genuine trade that the frame rate figure obscures completely.
The display specifications worth reading
Refresh rate, which caps how often the image can update.
Response time, which is how fast pixels change and which manufacturers quote using a favourable measurement method. Independent testing is more reliable than the specification.
Input lag, which is the display's own processing delay and which is rarely published by manufacturers but is measured by independent reviewers.
And variable refresh rate support, including the actual range over which it operates, which is frequently narrower than advertised.
Where diminishing returns arrive
An honest note about how much of this matters.
The difference between thirty and sixty frames per second is dramatic and perceptible by anybody.
Between sixty and a hundred and twenty it is clearly perceptible, particularly in fast motion.
Above that, the differences are real, smaller, and matter mainly in competitive play where reaction time is the point.
The same shape applies to latency. Reducing it from a hundred milliseconds to fifty is transformative. From twenty to fifteen is measurable and barely perceptible to most people.
Which suggests spending effort on eliminating the worst cases — the stutters, the display processing, the badly configured synchronisation — rather than on pushing an already high average higher.
Measuring it yourself
Since published figures rarely cover your specific setup.
Frame time graphs are available in most overlay tools and are more informative than the frame rate counter most people leave on.
End-to-end latency is harder, and there are now tools and hardware that measure it directly, plus features built into some drivers and displays that report an approximation.
The simplest useful test is comparative rather than absolute — change one setting, play the same section, and see whether it feels different. Perception is the thing being optimised, and it is a legitimate instrument.
The peripheral side
The part of the chain closest to the user and the cheapest to improve.
Input device polling rate determines how often a mouse or controller reports position, and higher rates reduce the delay at that stage by a small and measurable amount.
Wireless peripherals have improved to the point where the latency difference against wired is negligible for most purposes, which was not true a few years ago.
The larger variable is usually the display, which is why a good monitor does more for perceived responsiveness than an expensive mouse.