Nvidia’s RTX Neural Texture Compression Cuts VRAM Use by 96%

Nvidia’s new RTX Neural Texture Compression (NTC) is making waves in gaming and 3D applications, with early benchmarks showing an incredible 96% reduction in texture memory usage.

Tested by Compusemble on an RTX 4090 at 1440p and 4K resolutions, this new AI-driven compression tech could revolutionize game graphics—but at a slight performance cost.

How Does RTX NTC Work?

RTX NTC replaces traditional texture compression methods by using tensor cores to compress and decompress textures dynamically. It has two main modes:

1. “NTC transcoded to BCn” – Compresses textures when loaded.
2. “Inference on Sample” – Decompresses textures only when needed, saving even more memory.

Benchmark Results: Huge VRAM Savings

• At 1440p with DLSS, the “Inference on Sample” mode reduced texture memory from 272MB to just 11.37MB—a 95.8% reduction.
• At 4K, memory savings remained high, and even at native resolution with TAA anti-aliasing, performance increased compared to DLSS-based rendering.
• Cooperative Vectors, a feature that optimizes memory handling, boosted 4K FPS from 650 to 1,500 when enabled.

Performance Trade-offs

While VRAM usage plummets, there’s a slight performance hit—especially in “Inference on Sample” mode, where FPS dropped by around 100 FPS compared to traditional compression. However, the next-gen RTX 5090 may see better optimizations.

A Game-Changer for High-Resolution Graphics?

RTX NTC represents the biggest advancement in texture compression since the 1990s. It could allow games to use 4X higher-resolution textures without increasing VRAM demands.

Surprisingly, the minimum GPU requirement is an RTX 20-series, but testing suggests it may work on GTX 10-series, AMD Radeon RX 6000, and even Intel Arc GPUs—opening the door for widespread adoption, including console gaming.

There’s no release date yet, but Nvidia’s breakthrough in texture compression could redefine the way high-end games and 3D applications handle graphics.