In this lesson, you will learn how to drastically reduce your draw calls by offloading the rendering of hundreds of identical objects to the GPU in a single pass.
When you have a dense environment—like a forest of identical trees or a field of scattered crates—Unity’s standard rendering pipeline treats each object as an individual command sent from the CPU to the GPU. If you have 500 crates, that is 500 individual draw calls, which creates a massive CPU bottleneck.
GPU Instancing solves this by sending the mesh data to the GPU memory just once, then sending a small buffer containing the unique positions, rotations, and scales of those 500 instances. The GPU then "stamps" that mesh across all provided coordinates in one go.
Enabling GPU Instancing
To implement this, your material must be "GPU Instancing compatible." If you are using a built-in shader like Standard, simply select your material and check the Enable GPU Instancing box in the Inspector.
If you are writing a custom shader, you need to add the following preprocessor directives to your shader code:
// Inside your Pass block
#pragma multi_compile_instancing
// Within your vertex shader input/output structs
UNITY_INSTANCING_BUFFER_START(Props)
UNITY_DEFINE_INSTANCED_PROP(float4, _Color)
UNITY_INSTANCING_BUFFER_END(Props)
Constraints to Remember
For GPU Instancing to trigger, every object must share the exact same Mesh and the exact same Material. If even one object in the group uses a slightly different texture or a unique property that isn't handled via MaterialPropertyBlock, the GPU will break the batch and create a new draw call.
While Static Batching is great for immobile world geometry, GPU Instancing is superior for massive quantities of identical objects—even those that move—because it drastically lowers the CPU overhead required to describe the scene. By shifting this burden to the GPU, you reclaim significant CPU cycles for your gameplay logic and physics calculations.
Key Takeaway: GPU Instancing collapses hundreds of individual draw calls into a single batch, allowing the GPU to replicate identical meshes efficiently while keeping the CPU free for game logic.