AMD Reduces Game Size by Up to 70% with AI

Last Monday (8), the AMD made its innovative tool available to developers Neural Texture Block Compressionrevealed not Eurographics Symposium on Rendering 2024.

This compression technology driven by artificial intelligence it achieves reduce the size of textures in games by up to 70% without sacrificing quality and without the need to modify shaders.

Read too
High reputation: AMD has higher brand recognition than Intel, Samsung and Xbox
Did you know that AMD already had the chance to buy NVIDIA?

AMD’s approach uses a layered texturing strategy to encode all texture data into a single compressed block. This method significantly reduces the amount of data required to render images, as well as the space used in both storage and VRAM.

Image reconstruction is done through inference, making the process viable without changes to the code execution pipelining, which facilitates its adoption.

Typically, rendering images in games requires the GPU to communicate frequently with the storage device to load the required textures. In other words, there is a constant transfer of data from the HD ou SSD to VRAM as the player explores the scenarios.

With NBTC, rendering occurs in parallel with data decompression for texture inference. Although it adds steps to the image generation process and results in latencies of 27 to 50 milliseconds, this system has the advantage of keeping all textures in VRAM. This eliminates the need to access the storage system, compensating for the additional latency and allowing for a reduction in the installation size of games.

Now, with this technology, will triple AAA games take up less storage space?