Nvidia's new texture compression tech slashes VRAM usage by up to 95%

Nvidia's new AI-based texture compression drastically cuts VRAM usage by up to 95%.

: Nvidia's Neural Texture Compression uses AI to substantially reduce VRAM usage, with potential savings reaching 95.8%. The technology involves three modes: Reference Material, NTC Transcoded to BCn, and Inference on Sample, with varying impacts on VRAM. Tested on a GeForce RTX 4090 by Compusemble, it highlights possible frame rate impacts despite freeing VRAM. Cooperative vectors optimize AI workloads, enhancing real-time rendering techniques.

Nvidia has developed Neural Texture Compression to address VRAM limitations in modern GPUs, achieving up to 95.8% savings. Compusemble showcased this technology using a GeForce RTX 4090 GPU, running tests at 1440p and 4K resolutions.

The technology uses three modes: Reference Material, NTC Transcoded to BCn, and Inference on Sample, balancing disk and VRAM usage. While effective in reducing texture memory, it may affect frame rates, especially with higher resolutions and DLSS usage.

Nvidia's emphasis on AI-powered rendering suggests that newer GPUs could mitigate these performance drawbacks. Meanwhile, cooperative vectors in rendering pipelines facilitate efficient AI operations, enhancing overall game rendering capabilities.

Nvidianeural networksGraphics