The paper claims the new algorithm for texture compression is supposedly better than traditional block compression (BC) methods and other advanced compression techniques such as AVIF and JPEG-XL.
Dubbed neural texture compression (NTC), the algorithm uses a neural network designed for material textures. Nvidia boffins built several small neural networks optimised for each material. Textures compressed with NTC preserve a lot more detail while being significantly smaller than even these same textures compressed with BC techniques to a quarter of the original resolution.
Researchers explain the idea behind their approach is to compress all these maps along with their mipmap chain into a single file, then have them be decompressed in real-time with the same random access as traditional block texture compression...
NTC is not really a complete cure for cancer. It can introduce visual degradation at low bitrates. The Boffins said that there was mild blurring, some of the finer details disappeared and there was colour banding and shifts, and features leaking between texture channels.
Game artists won't be able to optimise textures in all the same ways they do today, for instance, by lowering the resolution of texture maps for less important objects or NPCs. Nvidia says all maps need to be the same size before compression, which is bound to complicate workflows.
Texture filtering is limited and there is potential for image flickering and other visual artifacts when using textures compressed through NTC.