GPU Instancing for Dense Game Worlds in Unity

When creating a game world that feels alive with dense environments, such as forests, cities or battlefields with hundreds of units, developers often encounter a major rendering bottleneck known as draw calls. Each time Unity instructs the GPU to render an object, it generates a draw call, which involves CPU overhead for setting up materials, textures, and render states.

On desktop hardware, thousands of draw calls per frame are generally manageable, but on mobile devices, this number often needs to be kept under 200 to maintain a stable 60 FPS. Domi Online, an MMORPG with interactive forests where every tree can be chopped down, faced this challenge by needing to render 10,000+ trees across the visible world with minimal draw calls on mid-range hardware while still allowing individual object interactions.

The naive approach of giving each tree its own GameObject with its own MeshRenderer and material leads to 5,000 draw calls for just the trees alone in a forest scene. This is far too costly for mobile hardware, typically the primary rendering bottleneck rather than polygon count. To address this issue, several techniques can be employed:

1. The SRP Batcher: Unity's Scriptable Render Pipeline (SRP) Batcher reduces the cost per draw call by keeping material data persistent on the GPU. Objects that share the same shader but may have different material instances can be batched efficiently. While it does not reduce the number of draw calls directly, it makes each draw call cheaper, significantly improving performance in dense environments.

2. GPU Instancing: This is the most effective method for rendering thousands of identical or near-identical objects. By issuing a single draw call that tells the GPU to render the same mesh several thousand times, with each instance having its own transform matrix, GPU Instancing drastically reduces draw call overhead. In Domi Online, GPU instancing reduced the number of draw calls from thousands to a few, achieving frame rates of 58 FPS with 5,000 instanced trees on mid-range Android devices.

3. LOD Strategies: Level of Detail (LOD) strategies help manage draw calls further by using lower-resolution meshes at greater distances. This technique ensures that distant objects are rendered with simpler geometry, which requires fewer resources and reduces the computational load without sacrificing visual quality.

4. Interactive-Object Swap Pattern: For objects that need to be interacted with individually, such as combat-ready trees, a swap pattern can be used to temporarily replace these objects with more detailed, individually rendered meshes only when they are close enough to the player. This approach balances visual fidelity with performance by using instancing for distant objects and detailed rendering for interactable ones.

Understanding and implementing these techniques - SRP Batcher, GPU instancing, LOD strategies, and the interactive-object swap pattern - can dramatically improve rendering performance in dense game worlds, making them both beautiful and functional without sacrificing frame rates on mobile hardware.