FPS vs frame time: what your GPU benchmark numbers mean

Frames per second (FPS) is the number of complete images, or frames, the device renders in one second. Every frame of a game or a 3D scene has to be calculated by the GPU: geometry is transformed, pixels are shaded and the result is handed to the display. The more frames per second, the smoother motion looks, up to the limit of the screen.

FPS is always an average over some window of time. A reading of 60 FPS means that 60 frames were completed during the last second; it says nothing about whether those frames arrived evenly.

Frame time is the time it takes to produce a single frame, usually measured in milliseconds (ms). Instead of counting frames in a second, it looks at each frame individually. That makes it the better tool for spotting the moments when rendering briefly struggles.

Lower frame time is better. A device that renders every frame in 8 ms is working faster than one that needs 16 ms per frame.

Because there are 1,000 milliseconds in a second, the conversion is simple arithmetic: frame time (ms) = 1000 ÷ FPS, and FPS = 1000 ÷ frame time (ms). Some common values:

• 30 FPS ≈ 33.3 ms per frame
• 60 FPS ≈ 16.7 ms per frame
• 90 FPS ≈ 11.1 ms per frame
• 120 FPS ≈ 8.3 ms per frame
• 144 FPS ≈ 6.9 ms per frame

The relationship is not linear. Going from 30 to 60 FPS cuts about 16.7 ms from every frame, while going from 60 to 120 FPS cuts only about 8.3 ms. That is why small FPS differences at high frame rates matter less than the same difference at low frame rates, and why frame time is often the fairer way to compare.

Imagine, purely as an illustration, a second in which a device renders 54 frames at 15 ms each (810 ms in total) and then one frame that takes 190 ms. That is 55 frames in one second, an average of 55 FPS, which sounds smooth. But that single 190 ms frame is a freeze you would clearly notice.

This is the main weakness of average FPS: it spreads a few bad moments across many good ones. Two devices can report the same average while one feels perfectly smooth and the other stutters. Looking at frame times, or at the lowest FPS reached during a run, brings those moments back into view.

Minimum FPS is the lowest frame rate recorded during a run. It points to the hardest moments: a demanding part of the scene, a background task or the device starting to throttle. A minimum close to the average suggests consistent performance; a minimum far below it suggests dips.

Maximum FPS is the highest frame rate recorded, often during the easiest moments of the run. On many devices rendering is synchronised with the display, so the maximum may sit at or near the screen's refresh rate rather than showing the GPU's full headroom.

The spread between minimum, average and maximum is often more informative than any single number. A narrow spread means steady performance. A wide spread means the experience varies, even if the average looks good.

A frame-time spike is a single frame, or a short run of frames, that takes much longer than its neighbours. Common causes include:

• Loading or preparing new work, such as compiling shaders or loading assets the first time they are needed.
• Background activity competing for the CPU, memory bandwidth or storage.
• Heat, when the system lowers GPU clocks. This usually appears as a gradual rise in frame time rather than a single spike.

A steadily rising frame-time trend together with a rising temperature is the classic sign of thermal throttling. Our guide to thermal throttling explains why it happens.

The MobiBench GPU test renders and animates a detailed 3D brain-hologram model for about 60 seconds. At the end it reports FPS, minimum, maximum and average FPS, frame time, temperature, peak temperature and a GPU score based on the measured rendering performance. A useful way to read them:

1. Start with the average FPS and frame time for the overall picture.
2. Compare the minimum FPS with the average to judge consistency.
3. Check the peak temperature against the starting temperature to see how much heat built up.
4. Repeat the test under the same conditions before comparing results. See How to run a fair phone benchmark.