Capture your frame time data
Use a capture tool such as CapFrameX, RTSS, or a game's built-in benchmark to record a representative gameplay segment. Export the raw frame time column in milliseconds or microseconds.
Free, privacy-first
Frametime Variance Analyzer
Turn raw frametime samples into jitter, spike, and percentile stability metrics: diagnose stutter that average FPS hides.
Input values
Results
Results update as you use the interface above.
How to Analyze Frame Time Quality
Use this flow when your average FPS looks fine but gameplay still feels uneven, or when you want to compare two hardware configurations on equal terms.
Paste and configure
Paste the raw frame time values into the input field and set the correct unit (ms or µs). Adjust the spike threshold if needed. 35% above the median is a reasonable default that filters minor jitter while flagging real stutter events.
Read the percentile results
Check median, P95, and P99 first. If P99 is more than twice your median, you have a significant stutter problem. Compare the 1% low proxy against your average FPS to measure the quality floor of your session.
Identify the source
High variance with low spike count points to systemic load variation: investigate CPU scheduling or GPU pipeline pressure. High spike count with moderate variance suggests discrete stall events such as shader compilation or thermal throttling.
Validate your fix
After changing a driver, setting, or hardware configuration, recapture under the same scenario and paste the new data. Compare variance and spike count directly, not just average FPS, to confirm the improvement.
Frequently Asked Questions
Why not rely only on average FPS?
Average FPS hides frame pacing instability. A session at 120 average FPS with high variance can feel worse than a stable 80 FPS. Frametime variance and percentile metrics expose stutter that the average completely conceals.
What is the 1% low proxy in this tool?
It is approximated from the 99th percentile frame time and converted to FPS. This gives a comparable floor metric to the 1% lows reported by tools like CapFrameX or MSI Afterburner, though it is an estimate rather than a precise match.
How many samples should I provide?
At least 50 samples is the minimum for stable percentile analysis. 100–200 samples from a representative gameplay segment give much better confidence, particularly for P99 and spike count where small sample sizes cause high statistical noise.
Are non-numeric tokens allowed?
Yes. Non-numeric tokens are skipped and counted so you can see how many entries in your data were invalid. A high non-numeric count may indicate a formatting issue with your export.
How should I use spike count?
Use it as a consistency signal alongside variance. Low variance but moderate spike count suggests occasional isolated stall events rather than systemic pacing problems: the two require different fixes.