Neural Texture Compression: NVIDIA and Intel Slash GPU VRAM Usage

Nvidia has introduced Neural Texture Compression (NTC), an AI-driven technology designed to significantly reduce the amount of video RAM (VRAM) required to render high-quality textures in games. The technology was detailed during sessions at GTC 2026, where the company demonstrated its ability to maintain visual fidelity while drastically lowering memory overhead.

The primary goal of NTC is to address the increasing VRAM demands of modern, photorealistic games. As game environments become more complex, the industry has historically relied on upscaling to manage hardware demands, but VRAM usage has continued to rise sharply.

Technical Implementation of Neural Texture Compression

Unlike traditional block-based compression techniques, which use formats such as BC5, BC6, or BC7 applied in 4×4 pixel formats, NTC utilizes small neural networks to unpack textures within a scene. Instead of storing the textures in a conventional format, the AI emulates the textures to produce the desired pixel format and appearance.

This approach allows developers to choose between two primary benefits depending on their goals: drastically reducing VRAM consumption or significantly enhancing the appearance of materials within the same memory budget.

Nvidia claims that this method can result in textures that look better than traditional methods, with the potential for up to 4x higher resolution in the final render. Because the workload is handled by AI technology, the company states these improvements can be achieved without a performance penalty.

Performance Demos and VRAM Reduction

During the GTC 2026 presentations, Nvidia provided specific examples to illustrate the efficiency of NTC. In a demo featuring a Tuscan Villa scene, the environment consumed 6.5 GB of VRAM when using standard block compression. By switching to Neural Texture Compression, the VRAM usage for the same scene dropped to 970 MB.

Nvidia reported that the resulting image appeared identical to the uncompressed version, demonstrating visual parity despite the memory reduction. The demo included both the exterior of the villa and an interior scene featuring detailed tableware to showcase the maintenance of material quality.

Another example showcased a flight helmet that originally contained 272 MB of uncompressed textures. While standard block compression reduced that size to 98 MB, NTC further reduced the memory footprint to 11 MB.

the company claims the technology can slash gaming GPU memory usage by up to 85% or reduce VRAM use by up to seven times in certain scenarios.

Industry Impact and Hardware Requirements

The implementation of NTC could make game installations more manageable by reducing the overall size of texture assets on disk. By freeing up substantial amounts of GPU memory, developers can incorporate more complex scenery and higher-quality materials without requiring users to possess excessive amounts of VRAM.

Nvidia has stated that the best graphics cards will be able to leverage this NTC technology to handle the AI-driven workload. This positions the technology as a way to maintain graphics fidelity while optimizing the efficiency of existing hardware.

The development of neural rendering is part of a broader trend in the industry to move away from traditional fixed-function compression toward AI-driven assets. This shift is mirrored by other players in the space, such as Intel, which has demonstrated Texture Set Neural Compression claiming up to 18x smaller texture sets.