There’s no denying that GPUs and VRAM are now being fed almost exclusively to AI, with little to nothing left for PC gamers around the world. Well, NVIDIA’s latest piece of gaming tech called the Neural Texture Compression (NTC), should solve that problem to a degree (emphasizing on the “should”, assuming they don’t lock it behind the latest gen of GeForce cards).
Instead of storing a pre-compressed texture, NTC stores a tiny neural network that knows how to rebuild the texture on the fly, done in a way that all material sets, such as color, normals, and roughness, get compressed together into one neural representation. Then, AI magic happens, and the texture is reconstructed in real time.
Because it understands how these texture layers relate to each other, it can preserve detail better than traditional methods, even allowing higher effective resolution at similar storage sizes.
Official demos showcased a 6.5GB of VRAM down to just 970MB with “visually no quality loss”, while another case sees hundreds of megabytes become just 11MB, which is honestly kind of insane when you think about it. Personally, I’d think that it’ll help out on the lower-end GPUs for better performance, while devs could push higher-quality textures without worrying.
But yeah, it’s not all upside. There is a trade-off here – you’re basically swapping memory usage for compute. Running that neural network costs GPU resources, and early tests show there can be some performance overhead depending on how it’s implemented. Also, this isn’t something you can just turn on in your GPU driver. Developers actually have to build it into games, similar to how DLSS works, so adoption is going to take time.
Right now, NTC is still in that early stage where it’s more of a “this is what’s possible” kind of tech. There’s already an SDK out for developers, and it works with modern APIs like DirectX and Vulkan, but you won’t see it in mainstream games just yet.
