AMD to Present Neural Texture Block Compression Technique – Reducing Game Sizes with Easy Game Integration

The GPU Open Twitter/X account revealed that AMD engineers S. Fujieda and T. Harada will present a neural texture block compression technique during next week's 35th Eurographics Symposium on Rendering. The session is scheduled to take place on July 2 at 3:30-3:45 PM local time at the Imperial College London, South Kensington, London, UK.

The main goal of this technique is to significantly reduce the ever-increasing size of games. Using a neural network, the textures (one of the main culprits) will be compressed to reduce the data size. AMD also promises 'unchanged runtime execution' that will help developers easily integrate the technique into their games.

Related Story AMD Might Re-Spec Ryzen 7 9700X “Zen 5” CPU With Higher 120W TDP To Outpace Its 7800X3D Chip

More details and possibly a full paper will be released next week. However, it's easy to imagine it won't be too different from NVIDIA's neural compression technique unveiled at SIGGRAPH 2023. Here's a basic overview of NVIDIA's technique:

The continuous advancement of photorealism in rendering is accompanied by a growth in texture data and, consequently, increasing storage and memory demands. To address this issue, we propose a novel neural compression technique specifically designed for material textures. We unlock two more levels of detail, i.e., 16× more texels, using low bitrate compression, with image quality that is better than advanced image compression techniques, such as AVIF and JPEG XL.

At the same time, our method allows on-demand, real-time decompression with random access similar to block texture compression on GPUs, enabling compression on disk and memory. The key idea behind our approach is compressing multiple material textures and their mipmap chains together, and using a small neural network, that is optimized for each material, to decompress them. Finally, we use a custom training implementation to achieve practical compression speeds, whose performance surpasses that of general