NT04 Recomp 1.0.0 brings Xbox NASCAR Thunder 2004 to desktop via WebAssembly

Boomhower34 released NT04 Recomp 1.0.0, a static recompilation of Xbox NASCAR Thunder 2004 running on Windows, macOS, and Linux.

olya_nro19 Nov 2026, 18:55· catalog:Decomp Buddy

NT04 Recomp, a static recompilation of EA SPORTS NASCAR Thunder 2004 from the original Xbox, has reached version 1.0.0. Built by boomhower34, it runs on Windows, macOS, and Linux using Node.js and a web browser, pairing a browser-based launcher with editors for rosters, careers, sponsors, tracks, and schedules.

The project translates x86 game code into C and compiles it to WebAssembly (recomp.mjs with clang). It pairs that recompiled core with a browser launcher and re-implementations of the Xbox pieces the game uses, including Direct3D over WebGL, DirectSound over Web Audio, threads, file system access, and input, plus fallback x86 emulation and JIT code when needed.

The release adds custom drivers, teams, and sponsors, a silly season career system with retirements, folding and returning teams, and contracts, custom tracks and schedule editing, steering wheel support, online races, render scaling, and optional picture upscaling. You need a .iso or .xiso disc image from a copy you own, as the repository contains no game files or game code.

To try a release build:

  1. Download the package for your OS from the project’s Releases page and extract it, as Node.js and the C compiler come bundled.
  2. Run Launch.bat on Windows, Launch.command on macOS, or launch.sh on Linux to open the launcher in your browser.
  3. Open the Setup tab under Game files, paste your disc image path, and click Extract.
  4. Under Game code, click Build game code and wait 5 to 15 minutes for the one-time build, then start the game from the Play tab.

Building from source requires Node.js 22 or newer plus LLVM 18 or newer (clang with the WebAssembly target and wasm-ld), with no npm packages required. Optional picture upscaling via upscale_textures.py needs Python 3 with numpy and pillow (plus PyTorch and spandrel for AI models).

Sources: github.com