Frame Time Explained — Why Milliseconds Beat FPS
FPS tells you how many; frame time tells you how it felt. This is the case for thinking in milliseconds — and how to read the graph that every serious performance tool now leads with.
The reciprocal that changes how you think
Frame time is simply 1,000 divided by frame rate: at 60 FPS each frame lasts 16.7 ms; at 120, 8.3 ms; at 240, 4.2 ms. The numbers look interchangeable, but they are not, because equal steps in FPS are wildly unequal steps in time. Climbing from 30 to 60 FPS saves 16.7 ms per frame — enormous. Climbing from 144 to 240 saves 2.8 ms. Same sport, different planets. This is why 'twice the FPS' is a meaningless boast without a starting point, and why honest comparisons — like the ones on this site's comparison pages — always quote the milliseconds saved.
Frame time also explains diminishing returns without any hand-waving about perception thresholds. Each doubling of frame rate halves the remaining frame time, so the absolute gain must shrink: there is less and less time left to save. Nothing mystical happens at any particular number; the curve simply flattens, and where you stop climbing it is a budget decision dressed as a sensory one.
Evenness: the second dimension FPS cannot show
A frame rate is one number summarising thousands of events; frame time keeps the events. Deliver frames at 8, 8, 8, 24, 8, 8 milliseconds and the average is a healthy-looking figure while the 24 is a felt hitch. This unevenness — pacing, judder, stutter, depending on its rhythm — is the dominant smoothness complaint on modern hardware precisely because raw rates have risen high enough that pacing is what remains. It is also why the graph matters more than the digits: a flat field of equal bars is smooth at almost any height, and a skyline is rough at almost any average.
Reading the graph on this site's tests takes seconds once you know the grammar. The cyan guide line marks your display's expected frame interval. Bars hugging it are health. Bars poking above it are late frames; the red ones took more than twice the interval and count as stutters. Clusters suggest a periodic cause — something on a timer; a rising slope across a long run suggests heat; random scatter suggests general load. The graph is a diagnostic instrument, not a decoration.
The table worth memorising
A few anchor pairs cover nearly every conversation: 30 FPS is 33.3 ms, the rhythm of older console games and broadcast sport replays; 60 is 16.7 ms, the desktop baseline; 90 is 11.1 ms, VR's comfort floor; 120 is 8.3 ms, the modern console and phone performance mode; 144 is 6.9 ms, the gaming-monitor standard; 165 is 6.1 ms; 240 is 4.2 ms, esports; 360 is 2.8 ms, the current ceiling of consumer panels. With those anchors, any claim in a forum or a review — 'this patch improved frame times by two milliseconds at 144' — becomes checkable arithmetic instead of vibes. That, in the end, is what frame time is for: it turns smoothness from an argument into a measurement, and it is why every tool on this site reports it next to the headline number.
Frequently asked questions
Is frame time more important than FPS?
They are the same information at the level of an average — reciprocals. Frame time wins where it matters: comparing improvements (milliseconds saved) and diagnosing unevenness, which a single FPS figure cannot express.
What is a good frame time?
One that matches your target rate and stays there: 16.7 ms for 60, 6.9 ms for 144, with minimal variation. Steadiness at a slightly slower rate usually feels better than a faster average with spikes.
Why do my frame times alternate long-short-long?
A classic pacing fault: the system produces frames faster than the display schedule for a beat, then waits — often a synchronisation or mixed-refresh interaction. The stutter test graph makes the pattern visible so you can correlate it with settings changes.
How does frame time relate to input lag?
A frame cannot show your input until it is presented, so frame time is a floor under one component of lag: at 60 FPS up to 16.7 ms of queuing; at 240, at most 4.2. It is one leg of the relay the input lag test describes.