What is 1% low FPS, and why does it matter more than the average?

An average frame rate can look healthy while a device stutters. The 1% low is the mean of the slowest frames — the number that matches how a device feels. How it is measured, what a gap from the average means, and why consistency is shown but not scored.

An average frame rate is a claim about how a device performs on the whole. The 1% low is a claim about its worst moments. Of the two, the second is the one that matches how the device feels in your hand, and it is the reason this test times every frame individually instead of counting them.

Why the average hides what you notice

Suppose a device draws sixty frames in a second, evenly — one every 16 milliseconds. It feels smooth. Now suppose it draws the same sixty frames, but one of them takes 400 milliseconds and the rest are crammed into what is left. The average is identical. The experience is not: the second device visibly froze. An average cannot tell those two apart, because it is the total divided by the count and both totals are the same.

That single stall is exactly the thing people mean when they say a device "lags". It is also exactly the thing an average is built to smooth over.

What the 1% low is

Take every frame that was drawn during the run, sort them from slowest to fastest, keep the slowest 1% of them, and average those. Express the result as a frame rate. That is the 1% low: the frame rate the device was delivering during its worst moments. A 400 ms stall that vanishes into the average dominates the 1% low, which is the point.

It is measured here from the render test — twenty thousand moving sprites drawn for three passes, with each frame's duration recorded. The frames from all three passes are pooled, because a percentile wants as many samples as it can get, and the first frame of each pass is excluded because it carries one-off start-up cost rather than steady-state delivery.

How to read the two numbers together

The useful thing is the gap. When the 1% low sits close beneath the average, frames arrive evenly and the device feels as fast as its average says. When the 1% low is far below the average, the device stutters — it is fast most of the time and then briefly not, and that "briefly not" is what you see.

This is why the power rating uses the 1% low rather than the average for the half of the score that is about smoothness. It is an absolute figure, so it falls both when a device stutters and when a device is simply slow, and it cannot be flattered by a good average.

Frame consistency, and why it is shown but not scored

The results screen also gives frame consistency: the 1% low as a share of the run's own average. It is a good description of feel — near 100% means the slowest frames barely differ from the typical one — and it is the only bar on the screen whose full mark is a real ceiling rather than an estimate, because the slowest frames cannot arrive faster than the average one.

It is deliberately kept out of the score, and the reason is worth knowing. A device that is uniformly slow delivers every frame late and delivers them all equally late, so its consistency is excellent. Scoring that would pay a device for being consistently terrible. The figure was tested by slowing a machine down artificially, and its consistency went up as it got slower. A ratio of a device to itself can describe it; it cannot rank it.

Worst frame and hitches

Two more figures come from the same frame timings. The worst frame is the single longest one in the run. The hitch count is how many frames took more than twice the run's median. One is the size of the worst stall; the other is how often stalls happened. A device with a high worst frame and few hitches had one bad moment. A device with many hitches has a pattern.

What none of this can tell you

Why. A stall can be the graphics driver, a background process, memory pressure, the operating system, or the device throttling under heat — and all of those look identical in a frame timing. The numbers say that the device stuttered and how badly. The guide on spotting throttling covers the one cause a benchmark can partly see, and the guide on what the score means covers how these figures feed the rating.

Related

• What does the score actually mean?Is 1,795 a good CPU score for a phone?
• Why is my new phone slower than my old one?My new phone feels slower than my old phone. Why?
• What makes a device throttle?What does it mean when a phone or laptop throttles?
• Is my phone throttling?How do I know if my phone is throttling?
• Why does the same device score differently?Why did my score change between runs?
• How to test a phone in a shopCan I test how fast a phone is before buying it?