Nvidia has spent years developing a new way to compress textures and save GPU memory. The technology is still in beta, but a newly released demo shows how an AI-based approach could help work around the VRAM limits of modern GPUs.
Nvidia Neural Texture Compression - A Fix for VRAM Limits
Nvidia's Neural Texture Compression (NTC) can save an enormous amount of VRAM when rendering complex 3D graphics, even though it has not yet been widely adopted. The YouTube channel Compusemble tested the technology through an official demo, which points to its potential for game developers in the near future.
According to Compusemble, RTX Neural Texture Compression uses a specialized neural network to compress and decompress textures on the fly. Nvidia's demo offers three different rendering modes:
- Reference Material: Textures stay in their original state with no NTC, which means high VRAM usage and large disk size.
- NTC Transcoded to BCn (the BCn block-compressed format): Textures are converted at load time, which sharply reduces storage size but saves only a moderate amount of VRAM.
- Inference on Sample: Textures are decompressed only when needed during rendering, delivering the largest savings in both VRAM and disk space.
Real-World Test Performance
https://www.youtube.com/watch?v=Z9VAloduypg
Compusemble ran the demo at 1440p and 4K, with both DLSS and TAA options. The results show that NTC not only cuts VRAM usage significantly but also affects frame rate (FPS). Specifically:
- At 1440p with DLSS, the NTC Transcoded to BCn mode reduces texture memory by 64%, from 272MB down to 98MB.
- The NTC Inference on Sample mode cuts texture memory to just 11.37MB, a 95.8% reduction compared with running without AI compression.
Impact on GPU Performance
The test ran on a GeForce RTX 4090, where DLSS and high resolutions add extra load on the Tensor Cores and affect performance depending on the settings. Newer GPU generations, however, may be able to optimize performance and minimize the drop in frame rate when using this technology.
Why Cooperative Vectors Matter for Rendering
The demo also highlights the role of Cooperative Vectors in modern rendering pipelines. As Microsoft has explained, Cooperative Vectors speed up real-time AI workloads by optimizing vector operations. These operations are important for training and fine-tuning AI models, and they also improve rendering performance in games.
With Nvidia investing heavily in AI-based rendering technologies like NTC, graphics are likely to see major improvements in VRAM optimization, giving developers more room to be creative in game design.
