1. Data Grid docs
  2. Developers

Performance

How the Data Grid stays responsive with a million records, and measured open, sort and search times against AG Grid Community.

A grid feels fast when the page never freezes: typing in the search box, scrolling and sorting answer at once, whatever the size of the data. The Data Grid is built for that, with the heavy work off the main thread and a DOM that stays the same size at ten rows or a million.

Open the large example: 1,000,000 records

How it stays fast

  • A Worker per grid. Files are read, and records are searched, filtered, sorted, grouped and totalled in a Worker. The main thread draws and handles input.
  • Columns in typed arrays. Records are held as one array per field, not one object per record. Text is stored once per distinct value.
  • No freeze while copying. Arrays of 4,096 or more objects are packed into columns in slices of a few milliseconds before they are sent to the Worker. Columns you pass with transferColumns: true move to it without a copy.
  • A bounded DOM. Only the rows and columns near the viewport are drawn.
  • Streaming files. XLSX and CSV files are read through slices of the file, never as one buffer.
  • Code on demand. The XLSX reader, the formula engine and the PDF writer load only when a file, a calculated field or an export needs them.
  • Server data in blocks. A server source is read in cached 100-row blocks, requested once scrolling settles and cancelled when the view changes.

Measured

On 18 November 2026 we compared 8sheets Data Grid with AG Grid Community 36.3.0. Both grids got the same generated records of 20 fields as JavaScript objects, in the minified build, in headless Chromium 131 (Playwright 1.49.1) at a 1200 × 800 viewport. Each figure is the median of five runs, in milliseconds; lower is better.

RecordsOpen (8sheets vs AG Grid)SortSearch
10,00070 vs 9623 vs 3633 vs 34 (a tie)
100,00088 vs 12334 vs 8533 vs 200

This measures these three local actions, on this data, on one machine. It is not an overall ranking of the two grids: other actions, data and settings give other results. Licence holders find the benchmark driver in the source download, src/table-view/dev/benchmark.mjs, to run it with their own data.

Getting the most out of it

  • Pass large data as columns ({ Units: Float64Array, … }) with transferColumns: true.
  • Give fields with types, so nothing has to be inferred.
  • Leave fields without formula or validator when you don't need them: plain records don't load the formula engine at all.
  • For more data than the browser should hold, use a server source; the page then holds at most 24 blocks of rows.
  • Keep cell renderers light: they run for every cell drawn while scrolling.