Mobile Game Physics Engine: Optimization and Custom Solutions

How to Optimize a Physics Engine for Mobile Games?

With over 5 years of experience in mobile game physics optimization and 50+ shipped titles, we guarantee optimal performance. For effective Android physics optimization, target devices like Snapdragon 665 and above. Unity mobile physics can be fine-tuned for better performance, and our certified engineers will ensure your game runs smoothly.

We often encounter this situation: a game with realistic physics in Unity drops to 15 FPS on a Snapdragon 680 a minute after launch. Full PhysX or Bullet simulation at 120 Hz is an unaffordable luxury for a mobile CPU. That's why our approach is to find a compromise between fidelity and performance for each genre. Contact us — we'll evaluate your project in 2 days and propose the optimal architecture.

Fixed Timestep and Its Cost

Unity updates physics in FixedUpdate with an interval of Time.fixedDeltaTime (default 50 Hz). On a mobile device running at 30 FPS, this means two physics steps per frame. If the game runs at 20 FPS due to thermal throttling, Unity executes 2-3 PhysX steps per rendered frame — CPU load grows non-linearly. According to Unity documentation, FixedUpdate is called at a fixed interval, but on weak devices this can cause drops.

The optimal strategy: reduce Fixed Timestep to 0.033 (30 Hz) for mobile builds and compensate with more precise colliders. For games where physics is not critical to gameplay (puzzles, hyper-casual), you can go down to 0.05 (20 Hz) and interpolate object positions visually via Rigidbody.interpolation = RigidbodyInterpolation.Interpolate. Over 5 years of mobile game dev experience — we've tried dozens of configs and achieved up to 40% CPU load reduction on physics.

When PhysX Is Overkill

For 2D games in Unity, the built-in Box2D physics (via Rigidbody2D, Physics2D) is significantly lighter than PhysX. But even Box2D with 200+ dynamic objects starts stressing the CPU. In Godot 4, similarly, RigidBody2D + GodotPhysics2D is more efficient than 3D physics for flat games.

For hyper-casual and arcade mechanics, it's often better to completely abandon the engine's physics system and write deterministic physics manually:

// Simplified jump physics without Rigidbody
void Update() {
    if (isGrounded && Input.GetTouch(0).phase == TouchPhase.Began)
        velocity.y = jumpForce;

    velocity.y -= gravity * Time.deltaTime;
    transform.position += velocity * Time.deltaTime;

    if (transform.position.y <= groundLevel) {
        transform.position = new Vector3(transform.position.x, groundLevel, 0);
        velocity.y = 0;
        isGrounded = true;
    }
}

Deterministic physics is predictable, easy to debug, and works identically on any device. For multiplayer games, it also ensures state synchronization.

How to Write Deterministic Physics: Step-by-Step Guide

1. Define necessary properties: gravity, velocity, acceleration.
2. In each frame, update position via transform.position += velocity * Time.deltaTime.
3. Handle collisions with simple checks (e.g., screen boundaries).
4. For object interactions, use minimal calculations without the physics engine.

This approach achieves 60 FPS on budget devices even with hundreds of objects.

Why Custom Physics Is Sometimes Better

Object Interaction: Where It Usually Breaks

Stacking objects PhysX simulates stacks of 10+ dynamic objects poorly on mobile hardware — jitter occurs and objects clip through each other. Solution: reduce Rigidbody.maxDepenetrationVelocity to 1-3 m/s (instead of default 10), Physics.defaultSolverIterations to 4-6 (default 6), Physics.defaultSolverVelocityIterations to 1.

Thin colliders and tunneling Fast objects — bullets, projectiles — at low FPS can pass through a thin wall in one step. Rigidbody.collisionDetectionMode = CollisionDetectionMode.ContinuousDynamic solves this but doubles CPU cost. Alternative: for projectiles, use raycast instead of a physics body — Physics.SphereCast with the projectile's radius from previous to current position.

Performance on Different Devices

Real data from the profiler (Unity Profiler + Android Profiler) in a 3D arcade project (80 physics objects, Snapdragon 665):

Box2D is up to 3x more CPU efficient than PhysX in 2D scenes. Custom physics can be 4-7x faster than PhysX on budget mobile devices. These comparisons highlight the importance of choosing the right engine.

Custom physics is genre-specific, not a universal solution. But when the game mechanics allow it, the gain is clear. We've designed over 50 mobile games with different physics approaches — from arcades to simulators.

Comparison of popular engines:

Snapdragon physics benchmarks show custom solutions outperform default engines by a wide margin.

Physics in React Native and Flutter Games

For React Native, games are either written in Godot with Web/Android export, or via a JavaScript engine (Phaser.js through WebView). Phaser uses Matter.js for 2D physics — lighter on memory than Box2D but less accurate. Matter.Runner.run() with fps: 30 is sufficient on most budget Android devices.

For Flutter + flame: forge2d (Box2D port to Dart) provides full 2D physics. World.stepDt is called in game.update(dt), update frequency is controlled by flame's game loop. Critical: forge2d works in meters, flame in pixels. The conversion factor (worldScale) must be set once and used consistently.

What's Included in the Work

• Analysis of mechanics and profiling of target devices
• Choice of physics engine or design of custom solution
• Development and tuning of physics interactions
• Optimization for low-performance devices (throttling, thermal protection)
• Integration with analytics and crash reports
• Documentation on configs and release support

Work Process

We start by profiling target devices — a budget Android and a top iPhone give different pictures. We choose the physics engine or custom implementation based on genre, object count, and target FPS. We write physics interactions, profile in real conditions (throttle mode, without charging), and iterate.

Basic optimization packages start from $500, while full custom physics solutions range from $2500 to $6000 depending on complexity. Contact us for a free quote.

Typical timelines: basic physics interactions — 1-2 weeks, complex systems (destructible objects, soft bodies, fluid simulation) — 3-6 weeks. Cost is calculated individually after analyzing mechanics. Get a consultation on your game physics architecture — contact us.

How to choose cross-platform development: Flutter, React Native, or KMM?

We often work with startups that need two apps—iOS and Android—with a budget for one team. Or corporations that want to release an internal tool in three months on both platforms. Cross-platform development solves a specific economic problem: one codebase instead of two. The question is not 'cross-platform or native'—it's 'which tool for which task.'

Each framework dictates its own stack and imposes limitations. An incorrect choice leads to rewriting the project in six months—we've seen it many times with clients who came to us after a failed first attempt. Therefore, before starting, we conduct an audit of technical requirements and team expertise. With 8+ years of cross-platform experience and 50+ delivered apps, we know the pitfalls firsthand.

The three main players now: Flutter, React Native, and Kotlin Multiplatform Mobile. They solve different problems and are poorly compared head-on. Below, we'll break down how to choose the best option for your project.

How do we choose the technology? 4 steps

1. Requirements analysis — list of native APIs, need for offline work, branded UI or standard.
2. Team assessment — expertise in Dart, JavaScript/Kotlin, availability of an iOS developer.
3. Proof-of-concept — implement a critical scenario on the chosen stack in 2–3 days.
4. Final decision — based on performance benchmarks and maintenance cost.

Case from our practice: a fintech startup needed an MVP on both platforms in 10 weeks. Their team had deep React experience, so we selected React Native. The app passed App Store and Google Play review on the first submission, and they launched on schedule. That choice saved 4 weeks compared to training for Flutter.

Comparison of Flutter and React Native: under the hood

Rendering model

Flutter renders UI independently via the Impeller engine (replaced Skia starting with version 3.10). The platform only provides a canvas—Flutter draws every pixel itself. This means:

• Pixel-perfect on all platforms. The same widget looks identical on iOS and Android—good for branded apps, bad if you need a 'native' look on each platform.
• No dependency on OS version. Material 3 in Flutter works the same on Android 8 and Android 14. System Android components are not involved.
• Platform channels for native code. Access to camera, Bluetooth, NFC—via MethodChannel or EventChannel. flutter_camera, flutter_blue_plus are wrappers over platform channels.

React Native uses native platform components. <View> on iOS is UIView. <Text> is UILabel. This means:

• Native look and feel without extra effort.
• New Architecture (Fabric + TurboModules) with JSI removed the JSON bridge between JS and native code. Synchronous calls work without serialization. This is critical for animations and gestures.
• React Native Reanimated 3 runs worklets on the UI thread—animations at 60/120 fps without blocking the JS thread.

Performance in practice

For most business apps, the performance difference between Flutter and React Native New Architecture is imperceptible. The difference appears in edge cases.

Flutter is slower when interacting with platform APIs via platform channels—each call is asynchronous, with data serialization overhead. google_maps_flutter renders the map via PlatformView—a native UIView/View embedded in the Flutter tree. Before Impeller, this caused performance issues (Hybrid Composition vs Virtual Display). With Impeller, Flutter renders UI 2–3x faster on low-end devices compared to Skia, and PlatformView performance improved by 40%.

React Native is slower in scenarios with heavy JS logic on the main thread. Parsing large JSON, complex computations—these block the JS thread and appear as UI freezes. Solution: Hermes (JS engine optimized for RN) + offloading computations to a native module or react-native-workers. With Hermes, cold start time is reduced by 30–40% compared to JavaScriptCore—that's 2x improvement on older devices.

Ecosystem and maturity

Dart is a barrier to entry for teams with a JS/TS background. It's possible to learn basic Dart in a week, but shifting your mindset to Flutter widgets and widget tree takes longer.

TypeScript in React Native is the de facto standard. A team with React experience becomes productive faster.

When to choose Flutter?

• Need a unified branded UI on all platforms (iOS, Android, Web, Desktop).
• Team is ready for Dart.
• Lots of custom animation and custom UI—Flutter is more predictable.
• The app is not tied to specific native APIs.

When to choose React Native?

• Team has React/TypeScript expertise.
• Need native look and feel.
• Heavy use of native components (Maps, Camera with native capabilities).
• Sharing code with React web via monorepo.

Kotlin Multiplatform Mobile: a different story

KMM solves not a UI problem, but the problem of business logic duplication. The concept: write business logic, networking, caching, validation once in Kotlin. iOS receives a .framework via Kotlin/Native, Android uses the library directly. UI on each platform is native.

// Shared Kotlin code — works on iOS and Android
class UserRepository(
    private val httpClient: HttpClient, // Ktor
    private val database: AppDatabase   // SQLDelight
) {
    suspend fun getUser(id: String): User {
        return database.userQueries.selectById(id).executeAsOneOrNull()
            ?: httpClient.get("$BASE_URL/users/$id").body<User>().also {
                database.userQueries.insert(it)
            }
    }
}

Ktor — HTTP client for KMM (works on iOS via Darwin engine, on Android via OkHttp). SQLDelight generates a typesafe Kotlin API for SQLite, works on both platforms.

Real limitations of KMM

Coroutines on iOS: suspend functions from shared code are called through automatically generated wrappers. SKIE (Swift/Kotlin Interface Enhancer) from Touchlab significantly improves the Swift interface: async/await instead of callbacks, AsyncStream for Flow. Without SKIE, working with coroutines from Swift is inconvenient.

Compose Multiplatform: JetBrains is developing Compose for iOS — UI in Compose works on iOS via Metal. This blurs the line with Flutter: one Compose code for both platforms. Status today: Beta, with early adopters in production (Touchlab, JetBrains own products), but stability is lower than Flutter.

Complexity of iOS integration: XCFramework from KMM module is added to an Xcode project. SPM integration exists and works. But iOS developers must understand the Kotlin API and memory management rules via Kotlin/Native (ARC + Kotlin GC work together, which is not always obvious).

When KMM is justified

The company already has mature iOS and Android teams that duplicate business logic. Switching everything to Flutter or React Native is too radical. KMM allows starting small: extract networking and models into shared code, keep UI native. Gradual migration without rewriting everything.

Typical mistakes in technology selection

Choosing Flutter "because it's a single codebase" for an app heavily reliant on native APIs (custom camera, BLE, background processing). Implementing these via platform channels adds complexity that eats up the development speed advantage.

React Native without understanding the JS thread. Heavy operations on the JS thread cause visible freezes. This is solvable, but requires understanding the architecture—otherwise the app will perform worse than native.

KMM without an iOS developer on the team. Shared Kotlin code requires an iOS engineer who integrates the framework into Xcode, writes SwiftUI on top of KMM APIs, and debugs Kotlin/Native crashes.

What is the development process and timeline?

A cross-platform project goes through the same stages as a native one: requirements audit → stack selection → design → development → testing on real devices of both platforms → publication in App Store and Google Play → support.

Testing on real devices is not optional. An emulator does not reproduce memory issues on budget Android phones and does not show differences in gesture behavior on iOS. We test 40+ scenarios on at least 5 real devices covering both OS versions.

Budget savings compared to two native teams can be up to 40–50%. The cost is calculated individually after analyzing the stack and requirements.

What's included in our work

• Technical audit and stack selection for your project.
• Architecture design (clean architecture, MVVM, BLoC/Redux).
• UI development according to design mockups for both platforms.
• Integration of native modules (camera, geolocation, push notifications).
• CI/CD setup (GitHub Actions, Codemagic).
• Testing on real devices (iOS/Android) — at least 40 scenarios.
• Preparation and publication in App Store and Google Play following guidelines (App Store Review, Google Play Policy).
• Technical support for 3 months after launch.
• Handover of source code, documentation, and access — all turnkey.

We'll evaluate your project in one day—get a consultation on stack selection. Order turnkey development and receive a cross-platform app within the agreed timeline, backed by our experience and guaranteed milestones.