1pct

Find out what held your last session back — or that nothing did.

Average frame rate hides the moments that ruin a session. Send one recording of a game and this tells you whether your machine was doing what it can, and what was in the way when it was not.

Your file is read, measured, and the raw frame times are discarded. No account, no email, nothing that identifies you or your machine. Nothing is published unless you choose to share the link.

A CapFrameX capture or a DxDiag report — either is welcome, and you do not have to say which. DxDiag is built into Windows and needs nothing installed.

What happens to your file, in full

Answer these two and this session joins the figures for that game. Leave them and you still get the full page.

Leaving either unanswered is a real answer, and here is what it costs: your page is built exactly the same and shows everything it would otherwise show, but the session is not pooled into any figure about this game. We would be comparing a 4K session against a 1080p one and calling it a population.

The capture file does not record either of these, and a page that compared a 4K session against a 1080p one would be comparing nothing. They are the only two things we ask you for.

No file yet? A DxDiag takes about a minute and needs nothing installed:

  1. Press Windows+R, type dxdiag, and press Enter.
  2. The first time, Windows may ask whether to check that your drivers are digitally signed. Either answer is fine; it does not change the report.
  3. In the window, click the button reading "Save All Information..." and choose where to put the file.

No file to hand? Look at what other people are running, with nothing to upload. Browse the population

A worked example: two recorded sessions, one laptop, both at 60 W.

Three numbers appear under each strip:

avg
Every frame in the capture, averaged. It is the number most sites publish, and it is the one that hides the slow moments.
1% low
The frame rate during the slowest one percent of frames. This is closer to what a rough patch actually feels like than the average is.
longest frame
The single longest frame in the capture. One frame this long is a visible hitch, whatever the average says.

Each bar is one block of the session, and its height is the share of that block's frames that ran over the stutter threshold: a taller bar is a rougher stretch of play.

The word above each strip names the pattern those heights make: clustered, where the rough blocks gather together, or continuous, where every block is rough. These two words are read off the blocks drawn here, not produced by a classifier: naming the shape in general needs a boundary fitted to many uploads, and those have not arrived, so a capture page states its blocks and declines the word.

ClusteredSilent Hill 2 Remake

block 1: 0.46% of 5,219 frames over the threshold, 60 sblock 2: 0.16% of 4,875 frames over the threshold, 60 sblock 3: 8.92% of 4,562 frames over the threshold, 60 sblock 4: 0.18% of 5,040 frames over the threshold, 60 sblock 5: 15.87% of 4,122 frames over the threshold, 60 sblock 6: 0.12% of 5,058 frames over the threshold, 60 sblock 7: 0.10% of 4,997 frames over the threshold, 60 sblock 8: 0.06% of 5,326 frames over the threshold, 60 sblock 9: 0.00% of 5,331 frames over the threshold, 60 sblock 10: 0.11% of 5,250 frames over the threshold, 60 s

avg 83.0 fps, 1% low 25.6, longest frame 1848 ms

n = one machine, one session, 49,780 frames: 600 s at 60 s blocks, a frame over 22.9 ms counts, 60 W median delivered

ContinuousResident Evil Requiem

block 1: 3.59% of 2,368 frames over the threshold, 15 sblock 2: 4.33% of 2,355 frames over the threshold, 15 sblock 3: 4.27% of 2,366 frames over the threshold, 15 sblock 4: 10.83% of 2,336 frames over the threshold, 15 sblock 5: 12.02% of 2,064 frames over the threshold, 15 sblock 6: 5.24% of 2,347 frames over the threshold, 15 sblock 7: 15.93% of 2,297 frames over the threshold, 15 sblock 8: 9.24% of 2,328 frames over the threshold, 15 s

avg 153.8 fps, 1% low 43.0, longest frame 174 ms

n = one machine, one session, 18,461 frames: 120 s at 15 s blocks, a frame over 12.8 ms counts, 60 W median delivered

This pair is fixed. It never changes and it counts nothing — uploads do not appear here. A file you send gets its own page, on a link only you hold.

Both sessions would pass any "can I run it" check, and by the usual measures of steadiness they are almost identical. The shape is the difference, and the shape is what this site measures. Each strip is scaled to its own tallest block, so read a block against its neighbours rather than across the pair.

The wattage and the thresholds above were read off the laptop while these two sessions were recorded. They are not in a capture file, so a page built from an upload states its blocks and says the delivered power was not recorded — these two are annotated from the bench they came from, and no upload is.

Real hardware reports counted

229,387

From the hardware attached to public Steam reviews of the games we hold: one report per review, so a machine whose owner reviewed several of these games is counted once for each. Not a survey, not a benchmark, not an estimate.

Games we hold

105

Every title with at least one machine counted, whether or not it clears the placement bar.

Have 32 GB of memory or more

25.0-88.5%

The range across 98 titles. No title sits at the average.

See the spread

corpus last grown 2026-09-05 01:11 UTC

Everything on this site