NVIDIA Shows Neural Texture Compression Cutting VRAM by 85% or Boosting Quality for the Same Budget
Wccftech • • technology
Key Points:
- NVIDIA reintroduced its Neural Texture Compression (NTC) technique during GTC 2026, highlighting its ability to efficiently store textures using machine learning by compressing textures into compact latent features reconstructed on demand by a neural network.
- NTC is deterministic, ensuring consistent texture reconstruction, and uses positional encoding to preserve fine details and repeating patterns, enabling accurate texture reproduction with high compression ratios.
- Compared to traditional formats like BCN, NTC offers significantly higher compression, supports complex assets with many material channels, and provides practical storage and bandwidth savings, which can reduce VRAM usage by up to 85% as demonstrated in a Tuscan Villa demo.
- The technology, available via a beta SDK since early 2026 but not yet adopted by game developers, is reportedly being considered by Sony to reduce PlayStation 6 game install sizes, benefiting from cross-hardware support including AMD and Intel platforms.
- The NTC SDK is accessible on GitHub, inviting developers to explore this advanced texture compression method that could improve game performance and distribution efficiency.
