NVIDIA's new bet on artificial intelligence applied to graphics has a name of its own: Neural Texture Compression (NTC)This technology, showcased in detail during the company's latest technical conferences, directly addresses one of the biggest bottlenecks in current gaming: the use of video memory or VRAM.
According to data provided by the firm, NTC is able to reduce memory consumption from approximately 6,5 GB up to about 970 MB in the same test scene, while maintaining virtually intact texture quality. We're talking about a reduction of nearly 85% in VRAM usage, something especially relevant for PC gamers with mid-range graphics cards or older systems, which are very common in Spain and the rest of Europe.
What exactly is Neural Texture Compression?
Neural Texture Compression is, in essence, a system of compression and decompression of textures based on neural networksInstead of storing the final texture directly in the GPU's memory, as with classic block formats (BC5, BC6, BC7 and similar), the graphics engine saves a compressed representation that a small neural network then reconstructs in real time.
This network learns to represent texels—the smallest unit of a texture—much more compactly than traditional methods. Thanks to this prior training, the GPU is able to recover the details of materials, surfaces and objects when it is necessary to render each frame, without needing to have all the original information permanently occupying VRAM.
From a developer's perspective, the idea is to replace raw textures or those stored in BCN format with these neural network versions. The change affects the internal graphics pipeline, but the ultimate goal is for the player to only notice the difference. Two things: less memory consumption and graphics at least at the same level., when not better for the same VRAM budget.
NVIDIA itself, in its technical talks at GDC and GTC, has framed NTC within a clear trend: integrating AI not only into visible tasks like image upscaling (as in the case of DLSS), but also into key stages of rendering that until now depended exclusively on fixed algorithms.
VRAM reduction: from 6,5 GB to 970 MB
The figure that has generated the most headlines is the demonstration with a complex scene, used as a reference in various presentations. In that test, a Tuscan-style villa with an abundance of detailed materials, dense geometry, and high-resolution textures It consumed around 6,5 GB of VRAM using standard BCn compression.
By activating Neural Texture Compression on the same assets, memory consumption dropped to around 970 MB of VRAM, maintaining virtually identical visual fidelityThe company also emphasized that it's not just about saving memory, but about using those savings to increase the level of detail if the game requires it.
In side-by-side comparisons, NVIDIA showed that, with the same VRAM budget, traditional compressed textures can generate visible artifacts, loss of sharpness, and degradation of materialsNTC, while retains more fine detail. In practice, this translates to cleaner surfaces, with less noise and banding, and better reflections and color transitions.
For the user, the potential impact is twofold: on the one hand, games that run better on graphics cards with 8 GB of VRAM or lessThis is very relevant in the European market, and on the other hand, titles that can increase texture resolution without raising the minimum memory requirements.
Paradigm shift from traditional texture compression
Most modern games use BCN formats to store textures directly in GPU memory. These formats divide the image into blocks and apply fixed-type compression techniquesThey are fast, highly optimized by hardware, and have been the standard on PCs and consoles for years.
However, they have a clear limit: to maintain a certain visual quality they need a minimum space per texelThis, in environments with 4K textures or a huge number of materials, easily fills up the VRAM. This is exacerbated in open-world games, dense urban environments, or productions with many cosmetic effects, which are very common in current releases.
Neural Texture Compression proposes a different approach. Instead of relying on a fixed compression scheme, it relies on machine learning models that have been pre-trained with large sets of textures and materials. These neural networks learn to encode and reconstruct visual patterns more efficiently than a conventional algorithm, especially when dealing with highly varied content.
In this way, what is stored is no longer the final texture, but a compressed representation that is then expanded on demand. The intensive use of the Tensor Cores present in GeForce RTX GPUs allows these inference operations to be performed in parallel with the rest of the graphics tasks, without overloading the main resources dedicated to rasterization and shading.
Impact on video games: lower requirements and higher quality
The direct consequence of all this is a possible significant reduction in minimum video memory requirements For games that adopt NTC. If textures, which usually occupy between 50% and 70% of the total VRAM in many titles, require significantly less space, there will be more room for the rest of the engine's elements.
This opens several interesting doors for European and Spanish studios that develop for PC and consoles, such as PlayStation 6Among the potential advantages, NVIDIA and various analysts point to the possibility of Use higher resolution textures on computers with less memorythus balancing the experience between players with cutting-edge hardware and those with more modest cards.
The size of the installations and patches also comes into play. By compressing assets more efficiently, it's possible that the games take up less disk space and that the updates weigh less, something that already worries both PC users with limited SSDs and those who play on consoles with restricted storage.
In the realm of asset streaming, so important in open worlds and titles that load data on demand, a smaller texture footprint can help to reduce bandwidth bottlenecksThis would result in fewer stutters, smoother loading times, and a more stable experience, even when the game is running from discs that aren't particularly fast.
Advantages for mid-range GPUs and laptop systems
One of the points that has generated the most interest in the community is the impact that Neural Texture Compression could have on graphics cards with 8 GB of VRAM or less, very widespread in the Spanish and European market, including some consoles such as Xbox Series XIn many recent releases, this type of GPU already encounters clear limitations when combining high resolutions and ultra-quality textures.
If a significant portion of the memory is freed up thanks to NTC, those same games could activate more aggressive texture adjustments without saturating the VRAMIn practical terms, this can translate into fewer sudden performance drops when loading new areas, less stuttering associated with memory usage spikes, and a more comfortable experience on 1440p or even 4K displays with balanced settings.
Portable systems, both gaming and light workstations, would also benefit. Although many modern laptops incorporate RTX GPUs, their The amount of video memory is usually more limited than its desktop counterparts. Having a technology that reduces the size of textures without degrading the image is especially interesting in this type of equipment.
For small or independent studios, common in the European scene, a reduction in VRAM requirements could help expand the potential user base without sacrificing a polished visual finish. This, in turn, aligns with a general industry trend towards seeking intelligent optimizations beyond the brute force of the hardware.
Neural Materials and other AI-based optimizations
Neural Texture Compression isn't alone. NVIDIA has also introduced the concept of Neural MaterialsThis is a complementary technique that aims to simplify how materials are processed within the graphics pipeline. Instead of handling many separate channels for each complex material, the information is condensed into a more compact representation that a small neural network decodes in real time.
In one of the technical demonstrations, it was shown how a set of materials that originally required 19 different channels could be reduced to just eight using this neural approach. According to the data provided, this simplification resulted in performance improvements ranging from 1,4 to 7,7 times at 1080p resolution, depending on the scene and model settings.
The key is that these networks are lightweight enough to be integrated directly into the shaders running on the GPU. Thanks to Tensor Cores, present since the GeForce RTX 20 series, the cost of these operations is kept under control, allowing Apply these optimizations millions of times per frame without blocking the rest of the rendering process.
Together, NTC and Neural Materials are aiming for a hybrid pipeline model, where traditional rasterization and ray tracing coexist with specific blocks of neural inferenceIn this scenario, AI not only improves the sharpness of the final image, but also handles structural tasks such as compression, shading, and memory management.
A graphic future shaped by AI
Although NVIDIA hasn't yet set a specific date for seeing Neural Texture Compression implemented on a large scale in commercial games, the demonstrations shown at events like GDC and GTC make it clear that the company wants this technology to be widely adopted. Be part of the next generational leap in graphics.
In the PC ecosystem, the adoption of APIs and extensions such as Cooperative Vectors in DirectX 12 This paves the way for these types of neural cores to run on hardware from other manufacturers as well. AMD has already announced support in future RDNA4 architectures, and Intel is working on similar initiatives for its graphics solutions, while companies like Sony strengthens visual computing.
If this cross-support is consolidated, neural texture compression could become a de facto standard in the industryThis benefits studios of all sizes. For European gamers, this could mean a longer lifespan for current GPUs, whose VRAM limitations would become less of a deciding factor in titles that incorporate these techniques.
In parallel, console manufacturers could leverage these solutions to further maximize the integrated memory of their systems, something especially interesting in long lifecycles where every optimization counts. Everything suggests that the next big graphics battle won't be fought solely on raw power, but also on... how the data that feeds each scene is managed and compressed.
NVIDIA's proposal with Neural Texture Compression and its associated technologies aligns with a shift in focus already evident in the industry: instead of endlessly increasing memory and computing power, the goal is for artificial intelligence to do more with less. If the figures seen in the demos—with VRAM reductions of around 85% and performance improvements in neural materials—translate to commercial games, players in Spain and across Europe could find themselves with visually more ambitious, better-optimized, and less memory-intensive titles, something that until very recently seemed difficult to achieve without sacrificing quality.
