Nvidia’s new NTC technology reduces VRAM requirements by up to 85%

Graphics cards are often limited by their memory. Nvidia’s new image compression technology could now solve this problem. The use of neural networks is expected to drastically reduce the VRAM requirement. A ray of hope for gamers during the storage crisis?

VRAM requirement drastically reduced

At the GTC 2026 developer conference, Nvidia gave deeper insights into the future of its so-called neural rendering technologies. However, the company initially received strong criticism for DLSS 5. Other functions were somewhat lost. This also applies to Neural Texture Compression, or NTC for short. The technology aims to curb the storage hunger of modern video games. This should please gamers, especially given the memory crisis that will probably last for a long time.

Instead of traditional block compression, NTC uses small neural networks to unpack textures in real time. This significantly reduces the load on the video memory. High-resolution textures in particular currently take up a lot of space in the memory of graphics cards. Nvidia’s approach shifts the workload to the GPU’s Tensor cores. Since the matrix accelerators work independently of the classic computing units, the basic performance of the graphics card remains unaffected.
Watch on YouTube

AI technology in practice

How Nvidia on YouTube demonstrated (from 11:35), the savings are impressively high. In a scene in a Tuscan villa, the VRAM requirement dropped from 6.5 gigabytes with standard compression to reportedly just 970 megabytes with NTC. The optical quality remains almost identical. In addition to textures, the company is also dedicated to the calculation of materials. At Neural Materials, a neural network calculates how light reacts to different surfaces. This usually requires complex mathematical formulas. According to Nvidia, the new approach speeds up rendering times by 7.7 times at a resolution of 1080p. The AI ​​models are trained in advance on game data, which means that no unwanted image errors should occur.

Historically, game data compression has been a constant part of game development. Back in the early 1990s, developers had to find creative ways to squeeze graphics onto small storage media. Formats such as S3 Texture Compression established themselves as industry standards and were later integrated into programming interfaces such as DirectX.

Still in development

It remains to be seen how quickly developers can incorporate these new tools into their game engines. If NTC becomes established, it could significantly ease the tense situation with limited graphics card memory. It also shows that machine learning can be used not only for pure image generation, but also for profound optimizations in the background. The new compression could curb the hardware hunger of future games.