Neural rendering is hotter than ever.
One after the other, at the Game Developers Conference (GDC) 2026, NVIDIA and Intel presented new technologies related to neural rendering. In San Francisco, teams from the two GPU giants competed in ingenuity to make the most of these techniques to reduce the size of textures used, in particular, in our video games. The stakes are manifold. By reducing the size of textures, we limit the need for video memory at a time when DRAM costs are reaching record levels. Smaller textures also mean data that is easier to manipulate and transfer more quickly. Finally, all in all, this opens the way to streaming gaming, which needs to reduce texture size considerably to be practicable.
At NVIDIA, it's called Neural Texture Compression (NTC) and is presented in the video above by Alexey Bekin (senior technology engineer and developer) and Shannon Woods (senior graphics program manager). The two engineers explain that neural rendering can be used in three ways. Firstly, at the end of the chain, in the manner of DLSS 5, to adjust game post-processing. It is also possible to adopt a completely generative approach to the image, leaving neural rendering in charge of almost everything. Finally, and this is the approach we're talking about today, you can intervene directly in the rendering management pipeline. The video above details the whole process, but let's remember that NTC delivers textures of virtually identical quality, while being 85% lighter. As a further exercise, for an equivalent size (970 MB in the example shown), the level of detail of NTC textures is far superior to that of the Block Compression n (BCn) generally used today.
However, NVIDIA is not alone in this field and, at the same GDC, Intel also demonstrated its Texture Set Neural Compression (TSNC) technology. Perhaps a little more technical than NVIDIA's, Intel's video also details the method employed, with the end result being the possibility of achieving a compression ratio up to 18x greater than that of BCn when TSNC Variant B is used. A less destructive Variant A can also be used to achieve a compression factor of 9 compared with BCn. As ever, when it comes to compression, you have to choose between the highest ratio and preserving maximum detail. The fact remains that both Intel and NVIDIA have adopted new techniques based on neural rendering, enabling impressive gains to be made without compromising visual quality. The future of video games, without a doubt.
