NVIDIA Neural Texture Compression: Reduce video memory usage without losing image quality

At the GTC conference, NVIDIA introduced Neural Texture Compression (NTC), designed to dramatically reduce GPU memory usage in modern games. As games become increasingly detailed, texture sizes grow, and hardware demands rise, making efficient optimization more important than ever.

Texture compression using neural networks

NTC replaces traditional block-based texture compression with small neural networks that process textures during rendering. This approach significantly reduces file sizes and decreases VRAM load without compromising visual quality, making game installations lighter and GPU workloads more efficient in real time.

Examples of VRAM reduction

In demonstrations, NVIDIA showcased a scene of a Tuscan villa: with standard compression, it occupied 6.5 GB of VRAM, while NTC reduced it to just 970 MB, with the image remaining virtually identical. Another example involved a pilot’s helmet: the original textures required 272 MB, block-based compression reduced them to 98 MB, and NTC brought it down to 11.37 MB — nearly 24 times smaller than the original size.

Neural Materials: faster rendering and enhanced lighting

Alongside texture compression, NVIDIA presented the Neural Materials concept, which allows neural networks to model light interactions with surfaces without complex physical calculations. Traditionally, GPUs compute lighting for all texture layers using BRDF models; now the network outputs final pixel shading directly, accounting for light and angle. In test scenes, this accelerated 1080p rendering up to 7.7 times without losing visual fidelity.

Efficient use of hardware accelerators

NTC leverages hardware accelerators — Tensor Cores in NVIDIA GPUs, XMX engines in Intel, and AI accelerators in AMD — enabling texture compression and lighting calculations without affecting baseline GPU performance.

Reliable and precise optimization

NTC is trained exclusively on the textures of the specific game, preventing generative artifacts or visual errors. This technology allows developers to create optimized, high-detail textures with accuracy and reliability, making it a safe and precise tool for modern game development.