NVIDIA Neural Texture Compression Enables 4x Higher Texture Resolution while Using 30% Less Memory


NVIDIA has developed a new compression technique called “Neural Texture Compression” (NTC) that delivers a level of quality that comes closer to what’s seen in uncompressed, reference assets, according to a new NVIDIA Research feature titled “Random-Access Neural Compression of Material Textures.” Per a comparison image that NVIDIA researchers shared, NTC provides a 4x higher resolution when compared to the BC (block compression) high method, which is pretty neat in itself, but what’s even more remarkable is that it’s able to accomplish this while using 30% less memory. More comparisons of the new compression method, which may or may not be exclusive to future GeForce GPUs, can be found here.

From an NVIDIA Research abstract:

The continuous advancement of photorealism in rendering is accompanied by a growth in texture data and, consequently, increasing storage and memory demands. To address this issue, we propose a novel neural compression technique specifically designed for material textures. We unlock two more levels of detail, i.e., 16X more texels, using low bitrate compression, with image quality that is better than advanced image compression techniques, such as AVIF and JPEG XL. At the same time, our method allows on-demand, real-time decompression with random access similar to block texture compression on GPUs, enabling compression on disk and memory.

The key idea behind our approach is compressing multiple material textures and their mipmap chains together, and using a small neural network, that is optimized for each material, to decompress them. Finally, we use a custom training implementation to achieve practical compression speeds, whose performance surpasses that of general frameworks, like PyTorch, by an order of magnitude.


Join the discussion for this post on our forums...

Recent News