RTX Spark Adds Support for Neural Texture Compression and Greatly Reduces VRAM Usage
Nvidia has officially brought its RTX Neural Texture Compression (NTC) technology to the Windows-on-Arm platform, with a particular focus on its upcoming RTX Spark systems. First shown during GTC 2026, the technology is designed to improve VRAM management in games and demanding applications by using artificial intelligence.
The main goal of NTC is to reduce VRAM usage by up to seven times without lowering image quality in games and applications. In recent demonstrations, Nvidia reduced scene memory usage from 6.5 GB to just 970 MB. Until now, the industry has relied on block compression formats such as BC5, BC6, and BC7, which are based on 4×4-pixel blocks. Nvidia replaces this method by training small neural networks that can recreate and render highly detailed textures while using only a fraction of the usual storage space.
RTX Spark Now Supports Nvidia’s Neural Texture Compression
The announcement comes alongside the first preview release of the native CUDA Toolkit for Windows-on-Arm. Nvidia wants to make sure the software ecosystem for RTX Spark is fully ready by the time the platform launches. RTX Spark systems will feature 20-core Arm processors, up to 128 GB of unified LPDDR5X memory, and Blackwell-based GPUs with between 5,120 and 6,144 CUDA cores.
The gaming industry is already responding to the new technology. Companies such as SEGA have started investing resources to optimize native AAA games for RTX Spark, taking advantage of the improved efficiency provided by NTC.
By reducing the need for high VRAM capacity and memory bandwidth, Nvidia is freeing up GPU resources for other real-time tasks while making it possible to deliver photorealistic graphics on portable and low-power devices.
