Your Average FPS Is Lying to You
Some links in this guide are affiliate links - we may earn a commission at no extra cost to you.
Someone posts a benchmark. Sixty frames per second, comfortably above the threshold everyone agrees is playable. Then they say the game feels awful, and the replies tell them their eyes are wrong.
Their eyes are not wrong. The number is just answering a question nobody asked.
What the number actually is
Frame rate is a rate. It counts how many frames arrived in the last second and reports the total. That is all it does.
Which means it cannot distinguish between these two seconds:
- Sixty frames, each arriving 16.7 milliseconds apart
- Fifty-six frames arriving 12ms apart, plus four that took 65ms each
Two runs that both average 60 FPS
Each bar is one frame. Taller means it took longer to arrive. Red bars are frames late enough that your eye catches them.
The averages are within a frame of each other. Nobody would describe these two runs the same way.
Both report 60 FPS. One is smooth and the other visibly hitches four times. The average has quietly thrown away the only information that distinguished them.
What 1% lows measure instead
Take every frame in a run and sort them by how long each one took. Look at the slowest 1%. Report that as a frame rate.
You now have a number describing your worst moments rather than your typical ones, and worst moments are what people actually notice. Nobody remembers the second where everything was fine.
The useful way to read it is as a ratio:
| 1% low vs average | What it feels like |
|---|---|
| Within 20% | Smooth. Nothing to fix. |
| 20% to 40% below | Noticeable roughness in busy scenes |
| Half or worse | Visible stutter, regardless of how high the average is |
That last row is the one worth internalising. A 120 FPS average with a 40 FPS 1% low feels worse than a locked, even 60. Higher average, worse experience. This is not a subtle effect and it is not subjective.
0.1% lows are the same idea with a finer sieve: the slowest frame in a thousand. Those catch individual events rather than general unevenness. A shader compiling for the first time. A texture being fetched from system memory. A new area streaming in.
Why bottlenecks appear here first
This is the part that connects to everything else on this site.
A component that cannot keep up does not fail evenly. It fails when the work spikes, and workload in games is spiky by nature: a corner turned, an explosion, twenty characters walking into frame at once.
Those are the moments a marginal CPU cannot prepare frames fast enough. Those are the moments a card short on video memory has to go fetch something. Your average absorbs them because they are brief. Your 1% low is made of them.
So a mismatched pair often looks fine on paper and plays badly, which is precisely the complaint that sends people to a bottleneck calculator in the first place. The mismatch was always visible. It was visible in the number nobody printed.
Two fingerprints worth learning to tell apart:
Repeated unevenness in dense scenes points at the processor. It arrives with crowds, physics, large battles - anything that raises simulation cost rather than rendering cost. Our CPU bottleneck explainer covers the mechanism.
Isolated large spikes when you turn to face something new point at memory pressure. That is the signature of a card fetching assets it has no room to hold, and it is the reason 8GB cards can benchmark well and still feel rough.
Measuring your own, in about five minutes
You do not need a capture card or a lab.
- Install a frame time logger. CapFrameX is free and purpose-built for this. MSI Afterburner with RivaTuner also works and you may already have it.
- Pick one repeatable 60 seconds. The same route through the same area, twice. Consistency matters more than length.
- Run it, then read the frame time graph, not the FPS number. You are looking for the shape: an even band with small variance, or a band with spikes sticking out of it.
- Compare the 1% low against the average using the ratio table above.
- Change exactly one thing and repeat. Texture quality is the best first test, because it isolates video memory better than any other setting.
That last step is where most people go wrong. Changing three settings at once tells you nothing about which one mattered.
What we would rather you took from this
Our opinion: the industry's habit of quoting a single average FPS figure has done real damage to how people buy hardware. It rewards parts that win a chart and punishes parts that feel better to play on, and it is why "just look at the benchmarks" is worse advice than it sounds.
Our admission: our own tools quote single numbers too. Our bottleneck calculator gives you a percentage, and our GPU scores are one figure per card. We do that because a single number is the only thing that fits in a comparison table, and because most people are choosing between two parts rather than diagnosing a machine they own. It is a simplification, and this page is the honest footnote to it. We have written before about how much a bottleneck calculator can legitimately know, including ours.
If you are diagnosing a machine you already own, the frame time graph beats every calculator on the internet, including this one.
Related reading
Found this useful? Pass it along:
Frequently Asked Questions
What are 1% lows in gaming?
Sort every frame in a benchmark run from slowest to fastest, take the slowest 1% of them, and report that as a frame rate. It is a measure of your worst moments rather than your typical ones. A 60 FPS average with a 58 FPS 1% low is a smooth experience. The same 60 FPS average with a 15 FPS 1% low stutters badly.
What is the difference between 1% lows and 0.1% lows?
Only the sample size. 1% lows cover the slowest one frame in a hundred, which is roughly what you feel as roughness. 0.1% lows cover the slowest one in a thousand, which is where individual hitches live - a shader compiling, a texture being fetched, a level streaming in. If your 1% low is healthy but your 0.1% low is terrible, you have occasional hitches rather than constant unevenness.
Why does my game stutter when my FPS is high?
Because frame rate is an average over a second and stutter is what happens inside that second. Sixty frames delivered evenly and sixty frames delivered in bursts produce the same number on an FPS counter and completely different experiences. Frame time graphs show the difference; FPS counters cannot.
What is a good 1% low?
Judge it as a ratio to your average, not as an absolute. Within about 20% of your average is smooth. Half your average or worse is visible stutter. A 120 FPS average with a 100 FPS 1% low is excellent; a 120 FPS average with a 40 FPS 1% low will feel worse than a locked 60.
Do 1% lows show a CPU or GPU bottleneck?
Both can hurt them, but they leave different fingerprints. A CPU limit tends to produce repeated unevenness in busy scenes, because the processor cannot prepare frames fast enough when there is a lot to simulate. A GPU running out of video memory produces isolated large spikes when new assets are needed. Sustained GPU load without memory pressure usually lowers the average smoothly and leaves the ratio intact.