Virtualization benchmark

The lab covers list, flow-grid, masonry, track-grid, and spatial layouts through equivalent public APIs. Within each family, adapters receive the same deterministic 100,000 items and the suite measures initial display, scroll response, and settling after inserts, moves, removals, and size changes.

Open the benchmark lab →

List

Comparison conditions

Uniform heights

Mixed natural heights

Warm layout readiness

Scroll

Insert

Move

Remove

Height change

Fixed height

Estimate provided

Height omitted

Uniform heights · Warm layout readiness · Fixed heightMax 14 ms

Warm layout readiness

Sectile Virtualv0.11.1 · Vue 3.5.223 rounds

Warm layout10.0 ms

10.0 ms

—

—

—

—

—

TanStack Virtualv3.14.10 · React 19.2.84 rounds

Warm layout6.5 ms

6.5 ms

—

—

—

—

—

react-windowv2.3.0 · React 19.2.84 rounds

Warm layout1.9 ms

1.9 ms

—

—

—

—

—

React Virtuosov4.18.12 · React 19.2.84 rounds

Warm layout14.0 ms

14.0 ms

—

—

—

—

—

react-virtualizedv9.22.6 · React 19.2.84 rounds

Warm layout2.1 ms

2.1 ms

—

—

—

—

—

Virtuav0.50.5 · React 19.2.83 rounds

Warm layout6.7 ms

6.7 ms

—

—

—

—

—

Vue Virtual Scrollerv3.0.5 · Vue 3.5.225 rounds

Warm layout13.9 ms

13.9 ms

—

—

—

—

—

Measurement criteria for this chart

Data

100,000 equal-height rows

Viewport

720 × 480px · 8-row overscan

Height input

Exact 72px supplied

Repetition

3–5 rounds with rotated library order

Completion

Correct total height and viewport geometry

Total height

Matches the measured total

Environment

Chrome 151 · Apple Silicon · macOS

Flow grid

Comparison conditions

Initial render

Scroll

Insert

Move

Remove

Resize

Fixed

Estimate provided

Automatic measurement

Initial render · FixedMax 20 ms

SetupFirst itemsStable layout

Sectile Virtualv0.11.1 · Vue 3.5.225 rounds

19.2 ms

19.2 ms

19.5 ms

React Virtuosov4.18.12 · React 19.2.8—

This condition is unsupported or has no measurement.

View raw measurements

Initial render and scrolling

Library	Input mode	Setup	First items	Stable layout	Scroll median	Scroll p95	Samples
Sectile Virtual	fixed	19.2 ms	19.2 ms	19.5 ms	3 ms	4 ms	100
Sectile Virtual	estimated	14.6 ms	14.6 ms	14.8 ms	2.8 ms	3.6 ms	100
Sectile Virtual	automatic	16.3 ms	16.3 ms	16.5 ms	1.9 ms	2.4 ms	100
React Virtuoso	automatic	0.9 ms	11.9 ms	12 ms	0.5 ms	0.7 ms	100

Layout mutations

Library	Input mode	Operation	Location	Scroll median	p95	Samples	Failed
Sectile Virtual	fixed	insert	start	8.1 ms	9.1 ms	30	0
Sectile Virtual	fixed	insert	middle	8.1 ms	9.2 ms	30	0
Sectile Virtual	fixed	insert	end	8.1 ms	9.1 ms	30	0
Sectile Virtual	fixed	move	start	8 ms	9.1 ms	30	0
Sectile Virtual	fixed	move	middle	8.1 ms	9.2 ms	30	0
Sectile Virtual	fixed	move	end	8 ms	9 ms	30	0
Sectile Virtual	fixed	remove	start	8.1 ms	9 ms	30	0
Sectile Virtual	fixed	remove	middle	8.1 ms	9.2 ms	50	0
Sectile Virtual	fixed	remove	end	8.1 ms	9.1 ms	30	0
Sectile Virtual	estimated	insert	start	8.2 ms	9.3 ms	30	0
Sectile Virtual	automatic	insert	start	8.2 ms	9.1 ms	30	0
React Virtuoso	automatic	insert	start	8.2 ms	9.2 ms	30	0
Sectile Virtual	estimated	insert	middle	7.9 ms	8.3 ms	30	0
Sectile Virtual	automatic	insert	middle	8 ms	8.9 ms	30	0
React Virtuoso	automatic	insert	middle	8.2 ms	8.8 ms	30	0
Sectile Virtual	estimated	insert	end	8.1 ms	9.1 ms	30	0
Sectile Virtual	automatic	insert	end	8.1 ms	8.5 ms	30	0
React Virtuoso	automatic	insert	end	8.2 ms	9.1 ms	30	0
Sectile Virtual	estimated	move	start	8.1 ms	9 ms	30	0
Sectile Virtual	automatic	move	start	8.2 ms	8.3 ms	30	0
React Virtuoso	automatic	move	start	8.1 ms	9.1 ms	30	0
Sectile Virtual	estimated	move	middle	8 ms	8.9 ms	30	0
Sectile Virtual	automatic	move	middle	8.1 ms	9 ms	30	0
React Virtuoso	automatic	move	middle	8.2 ms	9.1 ms	30	0
Sectile Virtual	estimated	move	end	8.1 ms	9 ms	30	0
Sectile Virtual	automatic	move	end	8.1 ms	9 ms	30	0
React Virtuoso	automatic	move	end	8.1 ms	9.2 ms	30	0
Sectile Virtual	estimated	remove	start	8.2 ms	8.4 ms	30	0
Sectile Virtual	automatic	remove	start	8.2 ms	8.3 ms	30	0
React Virtuoso	automatic	remove	start	8.2 ms	9.2 ms	30	0
Sectile Virtual	estimated	remove	middle	8.1 ms	8.8 ms	30	0
Sectile Virtual	automatic	remove	middle	8 ms	8.4 ms	30	0
React Virtuoso	automatic	remove	middle	8.2 ms	9.2 ms	30	0
Sectile Virtual	estimated	remove	end	8.1 ms	8.9 ms	30	0
Sectile Virtual	automatic	remove	end	8.1 ms	8.3 ms	30	0
React Virtuoso	automatic	remove	end	8.2 ms	8.9 ms	30	0
Sectile Virtual	estimated	resize	start	8 ms	8.8 ms	30	0
Sectile Virtual	automatic	resize	start	7.7 ms	9.3 ms	30	0
React Virtuoso	automatic	resize	start	8.2 ms	9.3 ms	30	0
Sectile Virtual	estimated	resize	middle	8.1 ms	9.2 ms	30	0
Sectile Virtual	automatic	resize	middle	8.1 ms	8.6 ms	30	0
React Virtuoso	automatic	resize	middle	8.2 ms	9.2 ms	30	0
Sectile Virtual	estimated	resize	end	8.1 ms	8.3 ms	30	0
Sectile Virtual	automatic	resize	end	8.2 ms	8.3 ms	30	0
React Virtuoso	automatic	resize	end	8.1 ms	8.7 ms	30	0

Masonry

Comparison conditions

Initial render

Scroll

Insert

Move

Remove

Resize

Fixed

Estimate provided

Automatic measurement

Initial render · FixedMax 15 ms

SetupFirst itemsStable layout

Sectile Virtualv0.11.1 · Vue 3.5.225 rounds

15.2 ms

15.2 ms

15.4 ms

TanStack Virtualv3.14.10 · React 19.2.8—

This condition is unsupported or has no measurement.

View raw measurements

Initial render and scrolling

Library	Input mode	Setup	First items	Stable layout	Scroll median	Scroll p95	Samples
Sectile Virtual	fixed	15.2 ms	15.2 ms	15.4 ms	2.8 ms	3.5 ms	100
Sectile Virtual	estimated	9.6 ms	9.6 ms	9.9 ms	6.4 ms	8.9 ms	100
Sectile Virtual	automatic	15.8 ms	15.8 ms	16.1 ms	6.1 ms	21 ms	100
TanStack Virtual	estimated	18 ms	18 ms	18.2 ms	0.4 ms	0.6 ms	100

Layout mutations

Library	Input mode	Operation	Location	Scroll median	p95	Samples	Failed
Sectile Virtual	fixed	insert	start	8.1 ms	8.3 ms	30	0
Sectile Virtual	fixed	insert	middle	8.2 ms	9.2 ms	30	0
Sectile Virtual	fixed	insert	end	8.1 ms	9 ms	30	0
Sectile Virtual	fixed	move	start	8.1 ms	8.9 ms	30	0
Sectile Virtual	fixed	move	middle	8.1 ms	9.1 ms	30	0
Sectile Virtual	fixed	move	end	8.1 ms	9.1 ms	30	0
Sectile Virtual	fixed	remove	start	8.2 ms	9.2 ms	30	0
Sectile Virtual	fixed	remove	middle	8.1 ms	9.2 ms	30	0
Sectile Virtual	fixed	remove	end	8.1 ms	8.2 ms	30	0
Sectile Virtual	estimated	insert	start	81.4 ms	134 ms	30	0
Sectile Virtual	automatic	insert	start	68.4 ms	138.9 ms	30	0
TanStack Virtual	estimated	insert	start	8.3 ms	9.6 ms	30	0
Sectile Virtual	estimated	insert	middle	43.2 ms	74.8 ms	40	0
Sectile Virtual	automatic	insert	middle	51.1 ms	85.3 ms	30	0
TanStack Virtual	estimated	insert	middle	8.6 ms	11.1 ms	30	0
Sectile Virtual	estimated	insert	end	7 ms	7.8 ms	30	0
Sectile Virtual	automatic	insert	end	6.8 ms	7.3 ms	30	0
TanStack Virtual	estimated	insert	end	8.2 ms	9.7 ms	30	0
Sectile Virtual	estimated	move	start	68.3 ms	136.1 ms	30	0
Sectile Virtual	automatic	move	start	67.8 ms	136.2 ms	30	0
TanStack Virtual	estimated	move	start	8.3 ms	9.8 ms	30	0
Sectile Virtual	estimated	move	middle	37.8 ms	72.7 ms	30	0
Sectile Virtual	automatic	move	middle	38.9 ms	72.7 ms	30	0
TanStack Virtual	estimated	move	middle	8.2 ms	10.3 ms	30	0
Sectile Virtual	estimated	move	end	6.9 ms	8.6 ms	30	0
Sectile Virtual	automatic	move	end	6.8 ms	7.2 ms	30	0
TanStack Virtual	estimated	move	end	8.3 ms	9.3 ms	30	0
Sectile Virtual	estimated	remove	start	73.6 ms	136.7 ms	30	0
Sectile Virtual	automatic	remove	start	71.9 ms	141.3 ms	30	0
TanStack Virtual	estimated	remove	start	8.3 ms	10.2 ms	30	0
Sectile Virtual	estimated	remove	middle	40.4 ms	76.5 ms	30	0
Sectile Virtual	automatic	remove	middle	40.1 ms	71.6 ms	30	0
TanStack Virtual	estimated	remove	middle	8.3 ms	9.3 ms	30	0
Sectile Virtual	estimated	remove	end	6.4 ms	6.8 ms	30	0
Sectile Virtual	automatic	remove	end	6.6 ms	7.2 ms	30	0
TanStack Virtual	estimated	remove	end	8.3 ms	9.3 ms	30	0
Sectile Virtual	estimated	resize	start	3.1 ms	4.4 ms	40	0
Sectile Virtual	automatic	resize	start	3.2 ms	3.7 ms	30	0
TanStack Virtual	estimated	resize	start	8.3 ms	9.6 ms	30	0
Sectile Virtual	estimated	resize	middle	3.1 ms	3.4 ms	30	0
Sectile Virtual	automatic	resize	middle	3 ms	5.9 ms	30	0
TanStack Virtual	estimated	resize	middle	8.2 ms	9.3 ms	30	0
Sectile Virtual	estimated	resize	end	3.2 ms	4.3 ms	30	0
Sectile Virtual	automatic	resize	end	3.1 ms	4.2 ms	30	0
TanStack Virtual	estimated	resize	end	8.2 ms	10.2 ms	30	0

Track grid

Comparison conditions

Initial render

Scroll

Insert

Move

Remove

Resize

Fixed

Estimate provided

Initial render · FixedMax 13 ms

SetupFirst itemsStable layout

Sectile Virtualv0.11.1 · Vue 3.5.225 rounds

12.6 ms

12.6 ms

13.2 ms

react-windowv2.3.0 · React 19.2.85 rounds

2.7 ms

2.7 ms

3.5 ms

Virtuav0.50.5 · React 19.2.8—

This condition is unsupported or has no measurement.

View raw measurements

Initial render and scrolling

Library	Input mode	Setup	First items	Stable layout	Scroll median	Scroll p95	Samples
Sectile Virtual	fixed	12.6 ms	12.6 ms	13.2 ms	8.7 ms	13.5 ms	100
react-window	fixed	2.7 ms	2.7 ms	3.5 ms	8 ms	9.5 ms	100
Virtua	estimated	Initial-render failure · geometry:item-37215

Layout mutations

Library	Input mode	Operation	Location	Scroll median	p95	Samples	Failed
Sectile Virtual	fixed	insert	start	11.4 ms	16.6 ms	30	0
react-window	fixed	insert	start	7.9 ms	8.3 ms	30	0
Sectile Virtual	fixed	insert	middle	12.4 ms	17.6 ms	30	0
react-window	fixed	insert	middle	7.7 ms	8 ms	30	0
Sectile Virtual	fixed	insert	end	9.7 ms	14.3 ms	30	0
react-window	fixed	insert	end	7.5 ms	8.4 ms	30	0
Sectile Virtual	fixed	move	start	9.1 ms	13.9 ms	30	0
react-window	fixed	move	start	7.8 ms	8.8 ms	50	0
Sectile Virtual	fixed	move	middle	10.1 ms	15.6 ms	30	0
react-window	fixed	move	middle	7.5 ms	7.9 ms	30	0
Sectile Virtual	fixed	move	end	10.2 ms	15.6 ms	30	0
react-window	fixed	move	end	7.5 ms	10.9 ms	50	0
Sectile Virtual	fixed	remove	start	11.3 ms	16.6 ms	30	0
react-window	fixed	remove	start	7.9 ms	8.2 ms	30	0
Sectile Virtual	fixed	remove	middle	12.4 ms	18.8 ms	30	0
react-window	fixed	remove	middle	7.4 ms	8.3 ms	30	0
Sectile Virtual	fixed	remove	end	10.2 ms	15.5 ms	30	0
react-window	fixed	remove	end	7.5 ms	8.7 ms	30	0
Sectile Virtual	fixed	resize	start	9.6 ms	14.6 ms	30	0
react-window	fixed	resize	start	7.9 ms	8.8 ms	30	0
Sectile Virtual	fixed	resize	middle	10.7 ms	16.1 ms	30	0
react-window	fixed	resize	middle	7.5 ms	8.3 ms	30	0
Sectile Virtual	fixed	resize	end	9.5 ms	14.5 ms	30	0
react-window	fixed	resize	end	7.5 ms	8.4 ms	30	0

Spatial

Comparison conditions

Initial render

Scroll

Insert

Move

Remove

Resize

Positioned

Initial render · PositionedMax 90 ms

SetupFirst itemsStable layout

Sectile Virtualv0.11.1 · Vue 3.5.225 rounds

90 ms

90 ms

90.4 ms

View raw measurements

Initial render and scrolling

Library	Input mode	Setup	First items	Stable layout	Scroll median	Scroll p95	Samples
Sectile Virtual	positioned	90 ms	90 ms	90.4 ms	7.2 ms	12 ms	100

Layout mutations

Library	Input mode	Operation	Location	Scroll median	p95	Samples	Failed
Sectile Virtual	positioned	insert	start	141 ms	168.5 ms	30	0
Sectile Virtual	positioned	insert	middle	100.9 ms	130.8 ms	30	0
Sectile Virtual	positioned	insert	end	8.2 ms	9.2 ms	30	0
Sectile Virtual	positioned	move	start	8.1 ms	9.2 ms	30	0
Sectile Virtual	positioned	move	middle	8 ms	9 ms	30	0
Sectile Virtual	positioned	move	end	8 ms	9.5 ms	30	0
Sectile Virtual	positioned	remove	start	139.5 ms	161.7 ms	30	0
Sectile Virtual	positioned	remove	middle	98.9 ms	130.7 ms	30	0
Sectile Virtual	positioned	remove	end	8 ms	8.5 ms	30	0
Sectile Virtual	positioned	resize	start	8.1 ms	9 ms	30	0
Sectile Virtual	positioned	resize	middle	8 ms	9 ms	30	0
Sectile Virtual	positioned	resize	end	8 ms	8.6 ms	30	0

Row profiles

The uniform profile isolates each library's base cost with identical 72px rows. The heterogeneous profile uses 256 deterministic combinations of titles, summaries, tags, and expanded details. Its measured row heights range from 71px to 159px.

The application does not calculate per-row heights. Each library measures the DOM it renders, while a separate calibration fixture supplies geometry only to the correctness validator. Visible rows are checked for order, overlap, gaps, viewport coverage, and anchor position. Total-height estimation error is reported separately for heterogeneous rows because unseen content has not been measured yet.

For two-dimensional families, exact modes require the full content extent and absolute item geometry. When a public API accepts exact per-item sizes but lazily derives their aggregate extent, absolute item geometry remains exact while the unseen total extent is provisional. Estimated and DOM-discovered modes instead require the requested data revision, finite provisional extents, correct rendered-item identity and size, and a non-empty viewport. Mutation completion requires the affected inserted, adjacent-moved, removed, or resized items to be observable in the rendered DOM. List and two-dimensional families share the same target-positioning and frame-settlement kernel; only their snapshot and oracle checks differ. Two-dimensional mutation samples retain both the first observer probe with correct geometry and the following verified-frame result. Probe cost is reported separately, while the existing frame result remains the conservative comparison score. This keeps unseen-item estimates and implementation-defined masonry lane choices out of the correctness score without accepting an outer framework commit as completed virtualizer work. Fixed flow-grid and masonry conditions use uniform items, and size-change scenarios run only where the public API supports changing item size.

Why height input is split into three modes

The suite measures exact heights, estimates, and omitted height input separately. Combining them would hide the cost difference between a fixed-size fast path and DOM measurement.

• Fixed: the application supplies the exact 72px height. This mode is available for the uniform profile.
• Estimated: the application supplies one common 72px estimate, then the library reads the actual DOM height.
• No height input: the application omits both an exact height and an estimate. Sectile renders an initial viewport-sized sample, measures those elements, and derives the remaining extent from those measurements.

Libraries that cannot start without a height or estimate are not given a synthetic value. The support table records the required input instead.

The height-change scenario expands one visible row to 96px. The browser DOM supplies the measured height; the exact 96px value is not passed back into the virtualization API.

Each adapter follows its library’s required integration. react-virtualized requires an explicit measurement-cache invalidation and layout recomputation call. The other adapters use APIs that observe DOM size changes automatically. That application-side work remains inside the timing boundary.

Collection mutation scenario

Insert, move, remove, and height-change operations run at three collection positions:

• start: the first row;
• middle: row 50,000 aligned with the viewport start;
• end: the final row visible at the bottom boundary.

Insert, remove, and height change affect one row at that position. Move swaps it with the adjacent row. Every sample starts from the same 100,000-row collection, and each scenario runs up to 50 times.

Timing starts with the update request and ends on the first frame where DOM order, row heights, total scroll height, and anchor position are all correct. Independent instances contribute batches of 5, 5, 10, 10, 10, and 10 samples; a stable median and p95 can stop a condition after 30 samples. Between samples, the runner restores and verifies the original collection; a failed restore discards that instance and mounts a fresh one. Clean and recovered samples are counted separately. A recovered sample retains both its recovery time and every observed correctness failure.

The settle bars cover every sample that reached a correct screen. Recovery bars use only samples that showed an incorrect screen before recovering. A failed condition is shown instead of a timing bar.

Failure criteria

Every frame is checked for missing or duplicate IDs, incorrect order or height, gaps, overlap, blank viewport regions, total scroll-height errors, and unexpected anchor movement. Any incorrect frame creates a correctness record. An incorrect layout that remains unchanged for at least 300ms and eight frames is a hard failure. A layout whose coordinates keep changing remains under observation for up to two seconds. Sectile failures carry fatal severity in the raw result.

Initial render and scrolling

Both row profiles render 100,000 rows in a 720 × 480px viewport. Text, CSS, input data, and the requested eight-row overscan target are shared. The no-height-input path starts from measured DOM rather than a library-wide numeric fallback.

Each condition first runs once in a fresh same-origin browsing context. That diagnostic records the first correct layout and the next browser presentation opportunity. It is excluded from the warm score. The main runner then completes an untimed warm-up mount and rotates library order across three to five measured rounds. Warm rendering is split into committed scroller output, first row output, and the first state with correct viewport geometry. A condition stops after three rounds when its cumulative median and p95 are stable; otherwise it continues through round five. Uniform rows also require the exact total scroll height. Heterogeneous rows retain total-height estimation error as a separate result.

Each round discards five warm-up scrolls and records the next 20. The harness changes scrollTop after a frame boundary. Timing starts when the browser begins delivering the native scroll event and ends immediately after the runner reads the resulting DOM geometry. The exact target row, contiguous row geometry, and viewport coverage are then validated against that snapshot outside the timed interval. Uniform rows also validate total scroll height.

The chart uses the conservative upper bound taken after geometry reads. Raw samples also retain the lower bound before those reads, probe cost, correctness-check count, round and sample number, MAD, and per-round ranges. This keeps validation work out of the score without hiding measurement uncertainty or a slow round.

The list-through-spatial observation was recorded on 2026-08-31 in Chrome 151 on Apple Silicon macOS from the clean production build of commit ce84a7b. Sectile completed every supported initial-render and mutation condition without a correctness failure. Absolute timings vary by machine and browser state; compare relative results together with correctness data.

The suite covers TanStack Virtual, react-window, React Virtuoso, react-virtualized, Virtua, and Vue Virtual Scroller. Runner code and committed JSON live in benchmarks/virtual-ecosystem.

pnpm --filter @sectile/benchmark-virtual-ecosystem dev

Isolated Sectile layout work

The browser suite includes framework and DOM work. This command isolates @sectile/virtual extent indexing, viewport queries, and layout-plan creation:

pnpm --filter @sectile/virtual benchmark

Observed on 2026-08-27 with Node 24.19.0, Apple Silicon, macOS:

Operation	Median time
Extent lookup pair, 100k	0.676 µs
Extent lookup pair, 1m	0.870 µs
Linear window reduction, 100k	1.237 µs
Linear materialized plan, 100k	7.694 µs
Linear changed 32-item measurement batch, 100k	2.857 µs
Sparse track-grid query, 100k regions	13.841 µs
Sparse track-grid changed 32-row batch	5.451 µs
Masonry query, 100k items / 8 lanes	21.873 µs
Spatial packed-tree query, 100k overlapping items	39.982 µs

Strategy	100k index construction
sparse track grid	45.742 ms
masonry, 8 lanes	16.467 ms
spatial packed tree	73.927 ms

The query viewport moves across each domain instead of repeatedly hitting one cached coordinate. The raw report also includes partitioned track-grid construction, pinned-region queries, and changed-track batches. Re-run before and after a change on the same machine when evaluating a regression.