What a 1% low FPS is

Your FPS counter says the game is fast, but it still feels choppy? That’s where the 1% low comes in. It’s the slow end of a capture, shown as a frame rate next to the average. Read the pair on the counter during a busy moment, not in a menu, and it tells you whether the scene was even, heavy, or hitching. Only then is it time to start the frame rate work.

I check these two numbers on an RTX 5080 and an AMD Ryzen 9 9950X3D2, playing on a 1080p 240 Hz and a 1440p 155 Hz monitor. The tool does the math. Your job is to read the pair for the moment you just played. You’ll find more on the guides index.

How the two numbers sit

How the two numbers sit

A gap Close together Both low

The average is ahead of the low. Replay the same moment. One change, then read both numbers again.

Use the same counter for both passes. A percentile from one program and an average of the tail from another are not a before and after.

Frametime is what was measured

CapFrameX’s explanation of performance metrics, by Taxxor, dated 31 May 2020, starts one step before the label. In-game benches, OCAT, FRAPS, PresentMon, and CapFrameX all print FPS. What they measured is the gap between pictures. That gap is the frametime, in milliseconds. FPS is the invert: FPS = 1000 / frametime, and frametime = 1000 / FPS. A long gap is a low frame rate. A short gap is a high one. The frame-time calculator is that invert. Type the average and the 1% low from the same counter to see both waits in milliseconds.

The average is the easy middle. The 1% low is some cut of the slow end. The capture is a list of waits. The frame rate is how those waits were printed.

A 60 FPS frame is about 16.7 milliseconds. A 144 FPS frame is about 6.9. A 165 FPS frame is about 6.1. A 240 FPS frame, the Overwatch screen, is about 4.2. One 16.7 in a run of 4.2s is the hitch. The average can still look like 240.

CapFrameX’s later post on displaying metrics as FPS says reviewers print the invert because a high-is-good bar sits next to the average. Shown as milliseconds, a 1% low would become a 1% high. The underlying list is still waits.

Three calculations, one label

CapFrameX’s same metrics post is blunt: compared with percentiles, there is no clear definition of how x% low values are to be calculated, so you have to know the tool before you compare two numbers. Then it describes three cuts that all get called a 1% low.

A percentile sorts the frames and takes a place in that list. The 1st percentile of the frame rates is the cutoff where 99% of the frames were faster. The worst 1% of frames sit outside that cutoff, so they do not pull the number down. CapFrameX’s warning is that this is a count of frames, not a share of time. In their example, a 20 second capture with a P1 of 45 can still hide four stalls of 500 milliseconds. Those four frames are a small share of the frames and a large share of the second. You were not above 45 for 99% of the time. You were above it for 99% of the frames.

The other common cut averages the slowest frames instead of stepping over them. NVIDIA FrameView’s user guide says 1% Low FPS takes the slowest 1% frames and averages them. The closer that number sits to Avg FPS, the more consistent the experience. FrameView added that column because the older 1% FPS percentile missed how bad the worst stutter was. CapFrameX calls the same mean x% low average. Because the outliers are inside the average, the result sits lower than the percentile cutoff. A single long frame still moves it.

The third cut is time. CapFrameX’s 1.5.3 notes, and the metrics post, say the 1% low can add the slowest frametimes until they sum to 1% of the capture. In the 20 second example, that sum is 200 milliseconds. The frame that crosses that sum is the number. A note in the same post, dated 8 October 2022, says version 1.7.0 keeps both: x% low average for the mean of the slowest frames, and x% low integral for the frame you were below during that share of the time. The integral is the one that really does mean you were slower than that rate for 1% of the capture. The percentile does not.

Read the tool before you read the number. FrameView’s 1% Low is the average of the slowest frames. FrameView’s 1% FPS is the cutoff. CapFrameX can be either of its two names. Mixing them and calling the difference a gain is how a setting gets credit for a change of formula.

What FrameView and AMD actually print

FrameView’s overlay treats three numbers as the key metrics: Avg FPS, 1% Low FPS, and Avg PC Latency. The summary file then adds more columns that people also call lows. 0.1% Low FPS averages the slowest 0.1% of frames. 1% FPS is the percentile that separates the slowest 1% of frame rates from the fastest 99%. 0.1% FPS and 5% FPS are the same idea at other cuts. Two columns in one CSV, both with a 1 in the name, are not a before and after.

PC Latency is the other key metric, and it is not a 1% low. FrameView describes it as the average time between the PC receiving input and the frame being sent to the display, in milliseconds. It needs a game with PCL Stats. It does not include the mouse or the panel. NVIDIA Reflex is the line that tries to shorten that queue. I keep Reflex on in Overwatch when the menu lists it. I still read the average and the low as a separate pair.

AMD Software: Adrenalin Edition, article DH3-038, lists Frame Rate, Frame Time, 99th% FPS, and Stutter Rate. There is no column named 1% low. I sit in front of an NVIDIA card, so I do not pretend Adrenalin is my overlay. A 99th percentile of FPS is a different sort. Game Advisor (DH3-028) adds a 95th percentile described as the longest time to process a single frame. That is a frametime cut.

ASUS ROG, in a 13 February 2026 guide by Whitson Gordon, uses the averaged definition: 1% low is the average of the slowest 1% of frames. That matches FrameView’s 1% Low, not CapFrameX’s integral. It is not a fourth formula.

Present versus displayed

A frame the game handed over is not always a frame the panel showed. FrameView’s 1.6 release notes say all FPS metrics (average, 1% Low, and the rest) are now calculated from MsBetweenDisplayChange, the wait from one displayed frame to the next. The older MsBetweenPresents numbers remain in the log as MinimumPresentFPS, 1%LowPresentFPS, and 0.1%LowPresentFPS. NVIDIA’s reason is that display change better matches what you saw. A present that never reached the screen is the wrong wait for a smoothness claim.

FrameView 2.0’s release notes go one step further. FPS now counts dropped and zero-display presents, so stutter, 1% lows, and 10% highs track the experience more closely. The same note says the average may read slightly higher than before. That is a change of counting, not a faster PC. If I compared a 1.6-style present log with a 2.0 display log and called the gap a driver win, I would be grading the tool.

NVIDIA’s FrameView page says the app leverages PresentMon for analytics. PresentMon’s console readme separates application start-to-start from display-to-display. Those two intervals can disagree when a frame is generated, dropped, or held by the compositor.

Microsoft’s Xbox Game Bar opens with the Windows logo key and G and can show performance metrics. Intel’s Windows 11 note puts FPS in that Performance section. Game Bar is a live rate. It is not a 1% low unless the widget you pinned says so. Use one counter.

The gap, in one capture

FrameView’s own sentence is the useful reading: the closer 1% Low is to Avg FPS, the more consistent the experience. A low near the average is even delivery. A low far under the average is an interruption in a run that was otherwise fine. Both numbers low together means the scene is simply asking for more than the machine is finishing.

Even delivery can still be a slow scene. A device that is late on every frame has a beautiful ratio, because the slow frames match each other. The ratio is a description. It is not a prize for being steadily late. If both numbers sit together at 70 on a 240 Hz panel, the picture is even and short of the refresh. That is the increase FPS job, one setting at a time, same moment twice.

Say your average is 144 but your 1% low is 60. The game will stutter even though the average looks great. A low that looks comfortable beside a modest average can be the stutter beside a much higher one. Read the gap you have, not a low borrowed from someone else’s average.

CapFrameX also prints a median, the 50th percentile. When a few long frames pull the average down, the median stays in the middle of the list. That is a check that the average was dragged, not a third grade of the PC.

0.1% lows and one bad spike

CapFrameX says a 0.1% low taken from two or three frames, in a scene that randomly spikes once, will not repeat, and cannot grade the hardware or the setting. You need more than 1,000 frames before a 0.1% cut even drops two or three values. A thirty-second menu is not that list. A busy minute is closer.

ASUS’s February 2026 page lists the stalls that show up as a low that falls away from the average: a CPU hitch on a physics spike, a shader compile while new assets load, and a RAM or VRAM miss that has to fetch from the drive. Those are one-off shapes until they happen every time you enter the same room. The method is two captures of the same moment. If the gap remains, it belongs to that moment. If it does not, you timed a compile, a Discord overlay, or a Windows update in the background.

A 0.1% low is a thinner slice of the same idea. Treat it as a hint that something rare happened, not as a second score. I do not chase a 0.1% number on a short Overwatch fight. I replay the fight.

What the low is not

It is not the click. Reflex, and FrameView’s PCL column, time the path from input to a frame leaving the PC. A capture can have a calm pair and a late click, or a hitching pair and a short queue. I turn Reflex on for the click. I still read the low for the hitch.

It is not min FPS. CapFrameX says min FPS is either the single worst frame, which wanders, or the worst one-second average, which hides micro-stutters. The 1st percentile replaced a lot of those mins because it ignores the one frame that a background app owned. Some in-game benches still print “min” and mean a 5th percentile. Ask which.

It is not a 99th percentile of FPS. A 99th percentile of frametimes is the slow end. A 99th percentile of FPS is the fast end. CapFrameX says reviewers who started on frametime charts kept saying P99 after they switched the bars to FPS. AMD’s 99th% FPS column is their label.

It is not proof the card is weak. A high average with a collapsing low is a stall. Both numbers low is a heavy scene. A number stuck on the refresh is a cap or a wait. Do not spend a 1% low on a limiter you typed.

Generated frames can raise the average. They are extra pictures. FrameView 2.0 reports a generation multiplier when NVIDIA Smooth Motion is active. I leave generation off in a competitive match.

What to do with the pair

• Read the average and the low from one counter, in the busy moment, not a menu. On my 1080p 240 Hz monitor that’s a Zarya fight or everything that wants more pixels.
• If they sit together and the rate is the one you wanted, leave the scene.
• If they sit together and both are short of the refresh, the scene is heavy. Change one setting and play that moment again.
• If the low falls away from the average on two passes, something in that moment is stalling. One change, then both numbers.
• Keep the formula. A percentile, an average of the slowest frames, and a time cut are three numbers. Only the same one can be a before and after.
• A 0.1% low is a thinner slice. A short capture makes it noisy.
• A cap a few under the refresh, when VRR is on, is how you stop the card from leaving the range. NVIDIA’s latency guide says G-SYNC plus VSync plus Reflex can cap below refresh for you. The cap is not the low. It can stop the card from drawing frames that become the hitch.

ASUS’s fix list matches the frame-rate guide: update the driver, cut a CPU-heavy setting if the processor pins, cut textures if video memory is full, enable the memory profile the kit shipped with. Change one thing. Replay the fight.

The average is allowed to look calm. The low is the slow end of the capture you just took, by whatever cut that counter uses. The next setting is for the scene that stays heavy, or stays uneven, after you have read the pair twice. More guides.

Common questions

What is 1% low FPS?

It's the frame rate of your slowest frames. Tools take the worst 1% of frames in a capture and report them as an FPS number, so you can see the stutters an average hides.

What is a good 1% low FPS?

A good 1% low sits close to your average FPS. If your average is 200 and your 1% low is 160, the game feels smooth; if the 1% low drops to 60, you'll feel hitches even though the average looks great.

Is 1% low the same as the 1st percentile?

Not quite. The 1st percentile is the cutoff frame that 99% of frames beat, while a 1% low average includes those worst frames, so it usually reads a bit lower. Only compare numbers that came from the same tool.

How is 1% low FPS calculated?

Most tools average the slowest 1% of frames in your capture. NVIDIA FrameView does it that way, while CapFrameX offers both an average and a time-based version, so check which method your tool uses.

Why are my 1% lows so much lower than my average FPS?

Because a handful of long frames barely move the average but you still feel them as stutter. Shader compiles, background apps and asset loading are the usual causes, so replay the same moment to see if the gap is consistent.

Should I worry about one bad 1% low result?

Not yet. One spike can come from a scene change, a shader compile or a background app. Capture the same moment again, and if the gap shows up both times, it's a real problem worth fixing.

What is 0.1% low FPS?

It's an even thinner slice of your slowest frames, the worst 0.1%. In a short capture that's only two or three frames, so it bounces a lot between runs; use it as a hint, not a verdict.

Is 1% low more important than average FPS?

They matter together. Average FPS tells you how fast the game runs, while 1% lows tell you how smooth it feels. A high average with collapsing lows still feels choppy.

How do I improve my 1% lows?

Start by closing background apps and capping your frame rate slightly below what your PC holds steadily. If lows are still far below the average, look for shader compiles, a setting that causes hitches, or a CPU bottleneck; if both numbers are low, lower your graphics settings.