Intel has showcased a new AI-based technology called Texture Set Neural Compression, or TSNC, that can compress game textures down to one-eighteenth their standard size while keeping visual quality close enough to the original that most players won't notice the difference. The technology was detailed this week by TechPowerUp, following Intel releasing a new video and technical breakdown of the system.
Game textures today are typically compressed using BCn formats - a family of block-based compression standards that have been the industry baseline for years. Intel's TSNC works by training a neural network on millions of BCn-compressed textures, creating an AI model that can encode and decode textures in a far more compact representation. The result loads faster, uses less VRAM, and can reduce both download sizes and on-disk footprints significantly.
Two Modes, Two Trade-offs
Intel is offering two compression variants. Variant A targets roughly 9x compression with minimal quality loss - measured at around a 5% drop using NVIDIA's FLIP image comparison tool. For most players and most textures, that's effectively imperceptible. Variant B pushes to the full 18x compression ceiling, with a 7% quality drop. That's more noticeable but still within the range that many developers may find acceptable for lower-priority assets, background geometry, or far-distance detail.
The performance side looks solid on Intel's own hardware. Benchmarks using a "Panther Lake" system with Arc B390 integrated graphics show the AI model generating the first texture pixel in roughly 0.194 nanoseconds - fast enough that there's no perceptible latency during rendering. The XMX AI acceleration cores built into Intel's Arc GPU lineup handle the decoding workload, which means the gains are most significant on Intel hardware but the technology isn't exclusive to it in principle.
Why This Matters
VRAM has become one of the main bottlenecks in modern PC gaming. Titles increasingly ship with multi-gigabyte texture packs, and mid-range cards with 8GB or 12GB of VRAM regularly struggle with high-resolution assets in demanding games. A technology that can compress those textures by 9x to 18x while preserving visual fidelity would meaningfully extend the useful life of mid-range hardware - and reduce the pressure on both developers and players to keep chasing higher-memory GPUs.
Download and install sizes are a separate pain point. Modern AAA titles routinely exceed 100GB, with a large share of that being texture data. Even partial adoption of TSNC by developers could knock that down considerably.
Intel isn't alone in this space. NVIDIA has its own Neural Texture Compression technology, and the general approach of using AI to outperform traditional block compression is gaining traction across the industry. What Intel is showing here is that the idea works on its own silicon, and works well.
Timeline
Intel says TSNC is currently targeting an alpha release later in 2026, with beta and stable releases to follow. No concrete dates have been announced. The technology needs to be integrated into game engines and adopted by developers before players see it in actual titles, which means the practical impact is still some distance away. But the underlying results are real, and for anyone who has watched their VRAM warning light up during a texture-heavy scene, the direction is the right one.
---------------
Author: Alan Ward
Seattle News Desk