Nitro Benchmarks

A react-native (or actually; Expo) app that compares react-native's Turbo Modules vs expo's Expo Modules vs @mrousavy's Nitro Modules.

About

This app specifically benchmarks the time it takes to call a synchronous function and return a result back to JS.

There are two synchronous functions per module:

• addNumbers(a: number, b: number): number
• addStrings(a: string, b: string): string

For each module, I chose the recommended language for iOS - so the TurboModule is implemented in Objective-C, and both the Expo Module and Nitro Module are implemented in Swift.

The functions are run 100.000 times after a short timeout to make sure the JS VM isn't busy with any other scheduled microtasks, and the total execution time is measured with performance.now().

About Nitro

Nitro is an upcoming library/toolkit I have been working on because I hit a lot of limitations with my libraries (react-native-vision-camera, react-native-mmkv, react-native-filament, and so on). I will use Nitro in my libraries to allow me to make them much faster than they already are, and to benefit from some fundamental differences such as it's instance-based approach. The Frame in VisionCamera, or MMKV in MMKV, or anything in Filament are all a C++ objects - I needed a better way of exposing those to JS, so I built Nitro.

I will make Nitro publicly available for everyone to build native C++, Swift or Kotlin modules with! 🔥

Results

iPhone 15 Pro (Release)

On my iPhone 15 Pro I got these numbers in a release build:

	ExpoModules	TurboModules	NitroModules
100.000x addNumbers(...)	434.85	115.86ms	7.27ms
100.000x addStrings(...)	429.53ms	179.02ms	29.94ms

• For addNumbers(...), Nitro Modules is 59x as fast as ExpoModules, and 15x as fast as TurboModules 🔥
• For addStrings(...), Nitro Modules is 13x as fast as ExpoModules, and 5x as fast as TurboModules 🔥

---

Mac Studio (Apple M1 Max)

On my Mac Studio (Apple M1 Max, 32GB RAM), I got these numbers in a release build ("designed for iPad"):

	ExpoModules	TurboModules	NitroModules
100.000x addNumbers(...)	764.31ms	195.12ms	42.32ms
100.000x addStrings(...)	773.00ms	308.65ms	138.43ms

Real-world

Note that these are just very specific benchmarks. They profile JS <-> Native function execution time (call, argument parsing, and result argument conversion). In an actual real world app, performance may vary.

Correctness

Note: I put these benchmarks together in ~1 hour. My focus is on making Nitro Modules as fast as possible, but also as powerful as possible, so this is where my time is dedicated at. If I made any mistakes in the benchmark, please let me know so I can update it asap!

I don't intend to put Turbo or Expo modules in bad light here, they are insanely cool tools to build native modules.