Nvidia shows neural compression can cut VRAM usage from 6.5GB to 970MB

Nvidia demonstrated Neural Texture Compression (NTC) cutting VRAM use from about 6.5GB to 970MB in its Tuscan Wheels demo while keeping image quality close to the original.

NTC preserved more detail than standard block-compressed textures at the same memory budget, enabling smaller installs and reduced bandwidth.

Neural Materials reduced a 19-channel material to eight channels and produced 1.4x–7.7x faster 1080p render times in one example.

Nvidia plans to embed small neural networks deeper in engines to decode textures, evaluate materials, and reduce memory traffic rather than relying solely on end-of-pipeline upscaling.