Intel Unveils Texture Set Neural Compression With Up to 18x Smaller Texture Sets

Intel has revealed new technical details about its Texture Set Neural Compression technology, known as TSNC, a cutting-edge approach to reducing the size of texture assets used in games and real-time graphics applications. The technology promises dramatic reductions in texture data without requiring artists to rebuild their entire asset pipelines.

TSNC comes in two distinct variants tailored to different use cases. Variant A targets higher visual fidelity, delivering an impressive 9x reduction in texture set size while maintaining quality suitable for demanding applications. Variant B pushes compression even further, achieving up to 18x smaller texture sets at the cost of some visual detail, making it ideal for scenarios where storage and memory bandwidth are the primary constraints.

The technology leverages neural network-based compression techniques to encode texture information far more efficiently than traditional methods such as BCn block compression. Rather than compressing individual textures, TSNC operates on entire texture sets simultaneously, allowing the neural model to exploit redundancies across multiple channels and maps to achieve its substantial size reductions.

Intel's announcement signals growing industry momentum behind AI-assisted graphics compression as developers face mounting pressure to manage ever-growing asset sizes in modern titles. If widely adopted, TSNC could significantly reduce game install sizes, ease GPU memory pressure, and open doors for high-fidelity content on bandwidth-constrained platforms and devices.