An Optimised Voxel Engine for the Nintendo GameCube

This is my final individual university project, focused on creating a voxel engine optimised for the Nintendo GameCube from scratch using its native Graphic API Nintendo GX. The main goal was to maximise performance and memory efficiency to render the highest number of chunks while maintaining a stable 60 FPS on the console’s limited hardware.

Optimisations in the Voxel Engine

To achieve a smooth 60 FPS while rendering as many chunks as possible, I focused on both performance and memory efficiency with these key optimisations:

🗃️

Batching

Reduced CPU overhead by combining multiple draw calls into one.

🚫👀

Occlusion Culling

First culled entire cubes, then refined it to cull individual faces, improving performance.

💾

Memory Optimisation

Used bitfields in structs to pack data efficiently, allowing more chunks to fit in memory.

📜🎨

Display Lists

Pre-recorded draw commands to reduce calls and improve FPS, at the cost of memory.

🎥

Frustum Culling

Rendered only chunks within the camera’s view, optimising performance.

📦

Dynamic Chunk Management

Loaded and unloaded chunks dynamically, reducing memory use and improving load times.

🌍

Efficient Terrain Generation

Optimised terrain generation to minimise redundant calculations and improve speed.

🎬

Efficient Animations

Optimised water movement and bone animations.

Distance vs Frustum Culling

Distance from Camera

Distance + Frustum Culling

Chunks loaded with frustum culling applied

Stages of Optimisation

Structure

All Stages

Stage 0: No Optimisations
Stage 1: Batching
Stage 2: Occlusion Culling L1 (Blocks — Game Loop)
Stage 3: Occlusion Culling L2 (Faces — Game Loop)
Stage 4: Occlusion Culling L3 (Blocks — Precalculated)
Stage 5: Occlusion Culling L4 (Faces — Precalculated)
Stage 6: Structs Level 1
Stage 7: Display List
Stage 8: Structs Level 2
Stage 9: Further Memory Usage
Stage 10: Achieve 60 FPS

Stages of Optimisation: Summary

Stage 0: No Optimisations

FPS: 15

N° Chunks: 9

Draw Calls: 20488

Free Memory: 17.785 KB

Stage 10: All Optimisations

FPS: 60

N° Chunks: 289

Draw Calls: ~100

Free Memory: 2.663 KB

What Knowledge Have I Acquired?

⚡

Optimisation

Mastered techniques like batching, frustum culling, and memory optimisations using bitfields to improve performance on constrained hardware.

🔧

Low-Level Game Development

Gained experience working directly with the DevKitPro SDK, understanding how to maximise hardware potential.

💾

Memory Management

Learned how to efficiently manage memory, balancing performance and memory usage.

📊

Performance Testing

Improved ability to track and analyse performance using tools like std::chrono to optimise in real-time.

🎮

Hardware Constraints

Developed a deep understanding of working within the limitations of older consoles and how to achieve optimal performance.

Memory Optimisation: Struct Sizes

Struct Cubito (Initial Version)

Struct CubeFace: 5 bytes

Struct Cubito: 38 bytes

Baseline version: no memory optimisation applied.

Structs Level 1

Struct CubeFace: 2 bytes

Struct Cubito: 16 bytes

Memory usage reduced by 57.89% in Cubito and 60% in CubeFace.

Structs Level 2

Struct CubeFace: 1 byte

Struct Cubito: 10 bytes

Maximum optimisation: 73.68% reduction in Cubito and 80% in CubeFace!

🚧 Post-TFG Improvements

After submitting my final project, I’ve continued working on the engine. Here are some of the features I’ve been adding:

✅✏️

Stencil Outline

Stencil-style outline on the Kirby model. The GameCube has no stencil buffer, so I simulated it using alpha compare.

🔲💡

Shadows

Shadow mapping implementation for directional lighting. Work in progress.

Sebastián Valdés

Sebastián Valdés Sánchez