The race to optimize resources in video games is enriched by a new chapter: texture compression via artificial intelligence (AI). AMD and NVIDIA, the two giants in the graphics card sector, are challenging each other with innovation to make games lighter and more accessible, without sacrificing visual quality.
AMD announced the presentation of a new neural compression technique at the upcoming Eurographics Symposium, while NVIDIA has already flexed its muscles at SIGGRAPH 2023 with its technology. But what does all this mean for players?
The GPU Open Twitter/X account revealed that AMD engineers S. Fujieda and T. Harada will present a Neural compression technique for texture blocks during the 35th Eurographics Rendering Symposium next week. The session is scheduled for July 2nd at 3:30-3:45pm local time at Imperial College London, South Kensington, London, UK.
The company’s goal is to significantly reduce the growing size of games, which is mainly due to the quality of textures. Textures are two-dimensional images that cover 3D models in video games, giving them detail, color, and realism. However, high-resolution textures take up a lot of storage space and memory, slowing down loading times and game performance.
Texture compression has been a technique used for years to reduce file sizes, but it often comes at the cost of visual quality. This is where AI comes in, promising to compress textures more efficiently, preserving detail and minimizing artifacts.
AMD and NVIDIA are developing different techniques, but both based on the use of neural networks to analyze and compress textures. The goal is to obtain a high compression ratio without compromising the quality of the final image.
For AMD more details and possibly a complete document will be released next week, while for NVIDIA’s neural compression technique presented at SIGGRAPH 2023 below you will find a basic overview directly from the manufacturer:
“The continued advancement of photorealism in rendering is accompanied by a growth in texture data and, consequently, increasing demands on storage and memory. To address this problem, we propose a new neural compression technique specifically designed for material textures. We unlock two additional levels of detail, or 16x more texels, using low-bitrate compression, resulting in better image quality than advanced image compression techniques such as AVIF and JPEG XL.
At the same time, our method enables real-time, random-access, on-demand decompression similar to block texture compression on GPUs, enabling both disk and memory compression. The key idea of our approach is to compress multiple material textures and their mipmap chains together and use a small neural network, optimized for each material, to decompress them. Finally, we use a custom training implementation to achieve practical compression rates, outperforming general frameworks such as PyTorch by an order of magnitude.”
#AMD #NVIDIA #Challenge #Lighter #Textures #Video #Games