PERFORMANCE PROFILE COMPARISON

See where the work moved.

Compare two Speedscope files. Follow the call paths. Keep the evidence.

Baseline

No file

Choose a profile or drop it hereSpeedscope JSON · up to 24 MiBProfile

Processed locally. No upload or account.

Current

No file

Choose a profile or drop it hereSpeedscope JSON · up to 24 MiBProfile

Choose comparable workloads and capture settings.

A ⇄ B

Two captures. One clear comparison.

Choose a baseline and a current profile above, or load the fictional checkout example.

Export other formats to Speedscope JSON in Speedscope ↗ first.

BASELINE WEIGHT

CURRENT WEIGHT

OBSERVED DIFFERENCE

FUNCTION IDENTITIES

CALL PATHS

Differential flame graph

↘ Decrease↗ Increase

All observed weight

FUNCTIONS

Find the largest changes

WeightShow

Function / source

How to read this comparison

Weight is the measured quantity in your files, not necessarily elapsed runtime. Sampled totals sum the supplied weights; evented totals cover the full declared interval, including inactive gaps. Time units convert to milliseconds. Share differences are percentage points using each profile’s own total.

Flat inclusive weight counts a function once per stack, even during recursion. The flame graph preserves recursive call paths. Exact matching can split a function after its source line moves; ignoring lines can merge distinct symbols. Paths are not stripped or remapped.

Two captures do not establish a causal regression or statistical significance. Keep workload, profiler settings, and recording conditions comparable.

Snapshots include the selected files in full, including all profiles, function names and paths. Share only artifacts you intend to disclose. Nothing is stored after you close this page unless you download it.