What’s new in Metal

New quantized tensor formats and support for scale factors allow you to easily express quantized data. Leverage native support in Metal performance primitives to efficiently compress weights and take advantage of the Neural Accelerators on the M5 Pro and M5 Max.

A redesigned temporal upscaler harnesses both the Neural Engine and Neural Accelerators on M5 Pro and M5 Max. MetalFX can now reconstruct fine detail from significantly lower render resolutions, allowing you to maintain fluid frame rates at higher-quality settings.

You can also simplify your port using the latest MetalFX framework. Process subrectangles for dynamic resolution, pass motion vectors to save preprocessing time, and handle postprocess effects with distortion fields.

A new companion repository on GitHub provides skills and sample code to help you use AI coding agents of your choice throughout the porting process, so you can produce higher-quality ports and ship polished games on Apple platforms.