Thursday, April 9

Intel neural compression tech significantly reduces VRAM usage while gaming


Major players in the PC gaming hardware space, along with Microsoft through DirectX, are continuing to invest heavily in neural rendering technologies aimed at improving both development workflows and gaming performance. Intel first introduced its Texture Set Neural Compression (TSNC) last year, and an updated version was recently showcased at GDC, with plans to release it later this year.

Similar to NVIDIA’s Neural Texture Compression (NTC), Intel’s approach uses AI to dramatically reduce the memory requirements of traditional textures that rely on block compression. By adopting this new method, TSNC can shrink texture data by up to 18 times compared to its original size.

Intel plans to offer TSNC as a standalone SDK that converts standard BC1-compressed textures (commonly used in games) into a format optimized for modern GPUs and even CPUs to decompress efficiently.

 

One notable advantage of this technology extends beyond real-time rendering. TSNC can also help reduce game installation and update sizes. The neural network stores compressed texture data and reconstructs it when needed, whether during installation, loading, streaming, or even at the pixel level during gameplay. This flexibility allows developers to prioritize reduced storage usage, lower memory bandwidth, or decreased VRAM consumption depending on their needs.

According to the presentation from Intel above, TSNC currently operates in two modes: Variant A and the more aggressive Variant B. Even with the higher compression setting, the visual quality drop is minimal – around 7% – while achieving significant reductions in data size.

The technology is designed to work best with Intel GPUs that support XMX cores, including upcoming Panther Lake chips with integrated Arc B-Series graphics. However, it also includes fallback support for non-Intel hardware and CPUs, ensuring broader compatibility rather than being limited to a single ecosystem.

We can expect this technology to be shipped later this year as an alpha version, with beta and full stable releases expected later on, though no concrete timelines have been provided. For the latest news on hardware launches and industry developments, be sure to follow our dedicated hardware coverage.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *