Use 16-bit precision in shaders

By default, GPUs use 32-bit precision. You can use 16-bit precision instead in GPU calculations, which has the following benefits on mobile platforms:

• ShadersA program that runs on the GPU. More info
  See in Glossary use less memory, bandwidth, and power.
• Calculations are faster. Using fewer bits can improve how the GPU allocates registers.

Create a 16-bit variable

To use 16-bit precision in a shader, use half when you declare a scalar, a vector, or a matrix. For example:

half _Glossiness;
half4 _Color;
half4x4 _Matrix;

To use 16-bit precision in a texture sampler, add half4 when you declare it. For example:

Texture2D<half4> _MainTex;

16-bit precision might not be enough for some shader calculations. This might cause visible errors such as color banding or stuttering geometry. To check for errors, run your project on a platform that supports half. If there are errors, use float instead.

A half variable is stored in buffers with a size and alignment of 32 bits.

Use 16-bit values on more platforms

By default, half has no effect on higher performance platforms in Unity, for example platforms that use MacOS. A half variable becomes a float, and the GPU uses 32-bit values for calculations.

To use 16-bit precision on more platforms, go to Edit > Project Settings > Player and set Shader Precision Model to Use full sampler precision by default, lower precision explicity declared. Unity then treats half in your HLSL code as the following:

• min16float for scalars
• float for texture samplers

When Unity compiles shaders, min16float becomes a platform data type that allows the GPU to use 16-bit precision for calculations if it supports them. For example:

• If you use DirectX 11 or 12, the variable remains a min16float.
• If you use OpenGL, the variable becomes mediump.
• If you use Vulkan, the variable becomes a RelaxedPrecision float.
• If you use Metal, the variable becomes a half, which has a size and alignment of 2 bytes instead of 4 bytes.

To override the Shader Precision Model setting for textures, add _half when you declare a texture sampler. For example Texture2D<half4> _MainTex.

Additional resources

• Scalar data types in the Microsoft HLSL documentation.
• HLSL data types
• Optimize shader runtime performance
• Writing shaders for different graphics APIs
• Fixed precision