Intel is developing a new technology that can significantly reduce the size of game textures, helping save storage space and VRAM.
This system works similarly to Nvidia’s neural texture compression and is designed to keep image quality close to the original while making files much smaller.
Intel says its system can shrink textures by up to 9 times in quality mode and up to 18 times in a more aggressive mode.
There will also be two versions of the technology. One is designed for Intel’s XMX hardware for better performance, while the other can run on regular CPUs and GPUs but may be slower.
The technology uses existing compression (BC1) along with advanced math to reduce texture sizes.
Instead of just compressing images normally, it uses trained data to keep textures looking similar even after they are reduced in size. An encoder compresses the textures, and a decoder restores them when needed.
A fallback version uses a simpler method (FMA), which works on more hardware but is slower.
Intel said developers can use this system in four main ways. One method is to compress textures before downloading, so games take less time to install and use less storage.
The other methods work during gameplay. These include loading textures as the game starts, streaming them while playing, and loading them only when needed without storing them in VRAM. This last option is useful for systems with limited memory.
The system has two modes: variant A and variant B.
Variant A focuses on keeping quality high. For example, large 4K textures can be reduced from 64MB to 10.7MB while keeping the same resolution. Other textures are slightly reduced in quality and size.
Variant B compresses more aggressively. It keeps one texture at full quality, while others are reduced in resolution and compressed further, going as low as 0.17MB.
In testing, Intel compared its system with older compression methods.
Variant A achieved over 9x compression, and variant B reached 18x, while the older method only achieved about 4.8x.
This puts Intel’s results close to Nvidia’s approach, especially in the more aggressive mode.
Intel said its solution can work on a wider range of hardware, including non-Intel graphics cards, which gives it broader compatibility compared to some competing technologies.