Benchmarks you can re-run

Every number on our benchmark page is one command away. Here is the time each runner adds on top of the tasks, on one 9,603-task graph:

Cold build overhead

vx6.68 s

Turborepo11.18 s

Nx3 min 12 s

Vite Task8.51 s

Nothing changed

vx559 ms

Turborepo689 ms

Nx12.07 s

Vite Task10.71 s

Cold build CPU

vx42.19 s

Turborepo1 min 19 s

Nx5 min 31 s

Vite Task24.17 s

The table

Runner	Cold build overhead	Nothing changed	Cold build CPU
vx	6.68 s	559 ms	42.19 s
Turborepo	11.18 s	689 ms	1 min 19 s
Nx	3 min 12 s	12.07 s	5 min 31 s
Vite Task	8.51 s	10.71 s	24.17 s

Synthetic monorepo, 1,601 projects, 9,603 tasks, 30 levels · build 300 ms, test and typecheck 150 ms, lint 75 ms, publish 30 ms · concurrency 10

Measured: vx 0.0.0 @ ee4c63a57 on 2026-11-12, Turborepo 2.11.10 on 2026-11-12, Nx 24.0.2 on 2026-11-12, Vite Task 1.4.0 on 2026-11-12; linux x64, 4 cores (AMD EPYC 7763 64-Core Processor), 16 GB.

Lower is better. A cold build’s tasks alone take 1 min 53 s with an ideal schedule. With nothing changed the ideal is 0, so all of that time is the runner.

The table is filled from the latest run when the site is built: each change to vx re-measures vx, and a new Turborepo, Nx or Vite Task pin re-measures that tool.

Re-run it

$ bun private/vx-bench/compare.ts 3

Every task is a sleep of a fixed length, so the numbers measure the runner, not a compiler. Each runner runs the graph cold, with nothing changed, and with outputs restored.

The site takes its rows from vx run @vzn/vx-bench#site.numbers: one task per runner, each on a machine of its own.

flowchart LR
  G[generate.ts: one graph] --> V[vx config]
  G --> T[turbo.json]
  G --> N[Nx targets]
  G --> VT[vite.config.ts]
  V & T & N & VT --> C[compare.ts: one runner at a time]

Rule	Why
each tool in its native config	Turborepo turbo.json, Nx nx:run-commands targets, Vite Task vite.config.ts
concurrency 10 for all	same parallelism
vx as the compiled binary, from a vx lock snapshot	what users install and CI runs
CI mode	as CI runs them
one runner at a time	no daemon idle-contends for CPU
no edit row	Nx needs its daemon for one, and CI doesn’t run it

Turbo and Nx run as they would in CI (CI=1: Nx’s daemon off, and Turbo uses none for turbo run).

Reading the columns

The first two columns are wall time. The third is the runner’s own CPU time, minus the tasks. A daemon that outlives the run isn’t counted, so Turborepo’s and Nx’s CPU are lower bounds.

Every row is on Benchmarks.

The method

Every change to vx’s warm path ships with a number measured like this: before and after interleaved, fastest of N, “before” built from its own git worktree.

If a number here doesn’t reproduce on your machine, please file a bug.

More at vznjs.github.io/vx.