Nvidia’s RTX Neural Texture Compression Achieves 96% Reduction in Memory Usage

Nvidia’s RTX Neural Texture Compression Achieves 96% Reduction in Memory Usage

Key Points

  • RTX NTC reduces texture memory usage by up to 96%, significantly improving storage efficiency.
  • Two modes were tested: “NTC transcoded to BCn” and “Inference on Sample,” which offered the most memory savings.
  • Performance impact is minimal sometimes, but “Inference on Sample” mode leads to slight FPS drops.
  • NTC is compatible with older GPUs, including RTX 20-series, GTX 10-series, AMD Radeon RX 6000, and Intel Arc A-series, hinting at broader adoption.

Nvidia’s RTX Neural Texture Compression (NTC) has been tested, revealing its potential to reduce memory texture sizes in 3D applications significantly. Benchmarked by Compusemble on an RTX 4090 at 1440p and 4K resolutions, NTC demonstrated an impressive 96% reduction in texture memory usage compared to traditional compression techniques.

Compusemble tested two modes of NTC: “NTC transcoded to BCn” and “Inference on Sample.” The first mode transcodes textures to BCn upon loading, while the latter only decompresses necessary texels for rendering, further minimizing memory usage.

At 1440p with DLSS enabled, the “NTC transcoded to BCn” mode reduced texture memory usage by 64%, from 272MB to 98MB. The “Inference on Sample” mode achieved an even more drastic reduction, bringing memory usage down to just 11.37MB—an astounding 95.8% decrease. However, performance saw a slight impact, particularly in the inference mode.

While NTC provides substantial memory savings, it comes at a performance cost. In the “NTC inference on sample” mode, frame rates dipped slightly, dropping from the mid-1,600 FPS range to mid-1,500 FPS. Despite this, 1% lows still performed better than traditional texture compression.

Switching from DLSS to TAA anti-aliasing led to a notable FPS boost, showing that NTC may heavily tax tensor cores, potentially bottlenecking shader cores. At 4K resolution, FPS performance dropped further, with “NTC transcoded to BCn” mode averaging 1,100 FPS and “Inference on Sample” mode just under 1,000 FPS.

RTX NTC has been in development for years and represents the first major upgrade in texture compression since the 1990s. Currently, in beta, the technology allows for up to four times higher-resolution textures than current GPUs can typically support. Surprisingly, the minimum hardware requirement for NTC is as low as an RTX 20-series GPU, with validation extending to GTX 10 series, AMD Radeon RX 6000, and Intel Arc GPUs—suggesting potential adoption beyond Nvidia’s RTX lineup.

EDITORIAL TEAM
EDITORIAL TEAM
TechGolly editorial team led by Al Mahmud Al Mamun. He worked as an Editor-in-Chief at a world-leading professional research Magazine. Rasel Hossain and Enamul Kabir are supporting as Managing Editor. Our team is intercorporate with technologists, researchers, and technology writers. We have substantial knowledge and background in Information Technology (IT), Artificial Intelligence (AI), and Embedded Technology.

Read More

We are highly passionate and dedicated to delivering our readers the latest information and insights into technology innovation and trends. Our mission is to help understand industry professionals and enthusiasts about the complexities of technology and the latest advancements.

Visits Count

Last month: 23565
This month: 12164 🟢Running

Company

Contact Us

Follow Us

TECHNOLOGY ARTICLES

SERVICES

COMPANY

CONTACT US

FOLLOW US