Feature Library

Open huge CSV and Excel files without freezing

2.63 CPU ms
A still of the Open huge CSV and Excel files without freezing demo
Still
The live demo starts when it’s in view.Drop a CSV or .xlsx, choose one, or make a 50 MB sample · download it back as .xlsx or CSV

Reads 64 MB+ CSV and .xlsx files in the browser a piece at a time.

What it costs a frame

2.63 ms
GPU
not used
CPU, median
2.63 ms
CPU, 95th pct
3.79 ms
Draw calls
0
Weight
Heavy

Measured 4 November 2026 on 13th Gen Intel(R) Core(TM) i9-13900H, Chrome 156, 1920×1080 at 1×. 600 frames × 3 rounds, median kept. What ran: Measured exercising itself: every frame feeds 64 kB of generated CSV (about 1,000 rows, semicolons and decimal commas) through the streaming parser, reads its numbers, updates the column profile and repaints the counts, all inside update. That is reading about 3.8 MB a second on the page’s own thread; off the bench, reading runs between frames or in a worker and only the repaint is per frame. Phones: tiers are being measured next.

On every machine it was measured on
MachineGPU msCPU msCPU p95
unknown GPU · 13th Gen Intel(R) Core(TM) i9-13900Hreference · measuring toolnot timed2.633.79
apple · Apple A19 Proin a browser tab · 16 November 2026not timed1.903.06
Intel Arc 140V (Xe2-LPG) · Intel Core Ultra 7 258Vin a browser tab · 15 November 2026not timed2.522.85

Bars against one 16.7 ms frame. A browser-tab run is timed in a visible tab without the tool’s scripted input, so read it as close, not exact.

What it needs

Renders with
DOM and CSSPlain DOM and CSS: every current browser.
Your browser
Runs hereNo WebGPU needed
Depends on
NothingNo other packages

What it’s for

  1. Importing customer spreadsheets
  2. Data tools and dashboards
  3. Previewing a file before upload
  4. Exporting tables to Excel
  5. Offline and privacy-first apps

Pitfalls, in brief

  1. Values cut from one big string keep the whole string alive in V8: the CSV reader copies each long cell (own), and the xlsx scanner decodes cells from bytes.
  2. Read numbers on purpose: “1,234” is a thousand and some in one locale and one and a bit in another. readCell treats a single comma before exactly three digits as a thousands mark; isPlainNumber leaves codes with leading zeros (007, 0123) as text.
  3. An .xlsx reads its shared strings first and keeps them all: a sheet with millions of unique strings needs that memory before the first row.
  4. A formula cell gives the value Excel last saved.
  5. Awaiting a stream is not yielding: a Blob’s stream through DecompressionStream can resolve every piece as a microtask, so a read never lets the page paint.
3 more in brief
  1. Rows posted from a worker are copied (structured clone).
  2. The writer makes plain zips: no zip64, so every part and the whole file under 4 GB.
  3. Excel opens UTF-8 CSV correctly only with a byte-order mark (csvBlob(rows, { bom: true })), and expects ; between cells where the decimal mark is a comma.
8 pitfalls in full, the hook-up guide, the module source and its tests come with the part.

Where it comes from

From SpaciousPro. Fully custom code.

PKWARE .ZIP File Format Specification (APPNOTE.TXT): the zip layout the reader and writer follow, zip64 records included

ECMA-376 Office Open XML: the SpreadsheetML parts: workbook, shared strings, sheets, styles

RFC 4180 (CSV): quoted cells, doubled quotes and line breaks inside quotes

CRC-32 (ISO 3309, as zip uses it): the reflected 0xEDB88320 polynomial, a 256-entry table

George Marsaglia, “Xorshift RNGs” (2003): xorshift32 makes the demo’s sample data

csvxlsxexcelzipstreamingDecompressionStreamCompressionStreamweb workerfile importexportno dependencies

Near it in the frame