(Credit: Intel)
Intel and Nvidia showed off their respective AI-powered texture-compression technologies over the weekend, demonstrating impressive reductions in VRAM use while maintaining texture quality, or even improving it, WCCFTech reports.
Although no game developers have used either technique in public releases yet, any VRAM reductions in games would be welcome news amid the memory crunch.
Nvidia's Neural Texture Compression technology debuted in May 2023, but Nvidia has refined it over the past few years, and the latest version is its most impressive yet. At GTC 2026, Nvidia explained how its technology uses machine learning to compress textures, capturing the most important visual information while massively reducing file size. When the game needs it, a small neural network running on the GPU recreates the texture. Crucially, it's deterministic, not generative, so the same resulting texture is recreated every time. In other words, this isn't AI slop, like some argue DLSS 5 is.
The system can work with complex assets that have different channel data—such as texture roughness or reflectivity—and lets textures fit more neatly into VRAM, allowing many more to be stored in a smaller footprint. This should also lead to smaller game installs, smaller patches, and faster downloads. Considering some games are now over 200GB, that alone could be worthwhile.
In one example, Nvidia cut the VRAM usage of a demo Tuscan Villa scene from 6.5GB to just 970MB, an 85% reduction.
Intel's demonstration of its Texture Set Neural Compression technique was equally impressive, TechPowerUp reports. Its AI-based neural network takes texture input data and compresses it, reducing it by up to 18x. That's the maximum efficiency for reducing VRAM and storage requirements, though there is some granularity.
Intel's Variant A system achieves compression ratios up to nine times those of the standard textures, while maintaining visual quality at a level where any changes are unnoticeable. Intel claims 5% visual quality drop on that version, while Variant B, using an up to 18 times compression ratio, suggests a 7% loss in visual quality, though both figures seem arbitrary and subjective.
Importantly, Intel also showed the technology working on Arc B390 integrated graphics, highlighting a response time of less than 0.2 nanoseconds, meaning it can be achieved even on low-end graphics chips without interrupting the feel of the game.
It may be some time before developers start adopting these technologies, but big reductions in install size, VRAM usage, and perhaps ultimately memory pricing would be very welcome.