8sheets Spreadsheet / Comparisons / SpreadJS / Method
How the measurements are made.
The speed figures on the comparison page come from opening the same generated workbooks in both components, in the same browser, on the same machine, five times each.
Set-up
Same browser, same machine, same files.
- Products. 8sheets 1.0.0, the full app (ribbon, formula bar and panels) mounted with
mountWorkbook, opening files withopenFile. SpreadJS 19.2.6 (@mescius/spread-sheetsand@mescius/spread-sheets-iofrom npm) in its evaluation mode, which adds a watermark and an “Evaluation Version” sheet, opening files with its documentedspread.import. - Browser. Headless Chromium 153 (Playwright 1.63), viewport 1400 × 900. A new browser process for every run, so every run is cold: no code cache, no HTTP cache.
- Machine. Intel Core i7-11700KF (16 threads), 30 GB of memory, Linux. A shared workstation; the two products were measured back to back.
- Server. Both pages from the same local server with the same headers.
- Runs. Every figure is the median of five runs. A run in which the browser lost a request to a local network glitch was repeated, for either product.
- Date. 16 November 2026.
Files
Generated workbooks, the same for both.
- 100,000 rows × 10 columns (large.xlsx, 7.5 MB): one million cells, three formula columns, so 300,000 formulas. The file asks for a full recalculation when it is opened.
- 1,000,000 rows × 20 columns (96 MB): numbers, text and decimals, no formulas. A 1,000,000 × 10 file (49 MB) was added when SpreadJS reported the larger one as an incorrect format.
- Recalculation file: 100,000 formulas of the form
=J2*$L$1, all depending on the one cell L1.
Tests
What each figure measures.
Each timer starts when the page, its code and an empty workbook are ready and the file is in memory, and stops when the last row’s cell reads the expected value and the frame showing it has been drawn.
- Open. By default SpreadJS shows the values saved in the file and does not recalculate them; we checked this with a copy whose saved last value was wrong. 8sheets recalculates on open, as the file asks. The published pair therefore has SpreadJS with
fullRecalc: true: 4.52 s against 0.90 s. With its default settings SpreadJS took 3.46 s; in its incremental open mode, 5.52 s. - Longest freeze. The longest task on the page’s main thread during the open, from the browser’s Long Tasks API, cross-checked by the largest gap between animation frames. 8sheets reads the file in a background thread and had no task over 50 ms; the largest gap between frames was 17 ms.
- Memory. Memory of the whole browser tab process after opening the 100,000-row file and two forced garbage collections. It includes the 8sheets background thread and its typed arrays, so it is the fair figure; the JavaScript heap alone would favour 8sheets.
- Recalculation. L1 changed from 1 to 2, timed until all 100,000 dependents are current. SpreadJS recalculates inside
setValue, blocking the page for that time; 8sheets calculates in short slices. Later edits: 26 ms against 405 ms. - Download. Every JavaScript response from page load to the end of an open, including the background thread’s scripts, compressed with gzip. 8sheets loads its whole app; SpreadJS’s figure is its grid and import plugin, without its separately sold ribbon. The 8sheets figure, 1,603 kB, is from the deployed build, which is minified and loads rarely used parts (saving, encryption, some dialogs) on demand; the readable source served before that weighed 2,740 kB.
- Scrolling. Both held 60 frames a second while scrolling 2,000 rows. The headless browser can’t show the difference in headroom, so this row isn’t on the comparison page.
Repeat it. The scripts run in containers against the public npm packages. Write to hi@8sheets.com and we’ll send them, or open the 100,000-row file and read the time in the status bar.