NVIDIA dropped one of the most interesting pieces of tech news at GTC 2026 on April 4, and it has nothing to do with a new GPU. During a session titled “Introduction to Neural Rendering,” the company demonstrated Neural Texture Compression, NTC, a technology capable of reducing texture memory usage by 85% while maintaining nearly identical image quality. The demo spoke for itself: a Tuscan Villa scene that consumed 6.5 GB of VRAM using traditional BCn block compression shrank down to just 970 MB with NTC active. Same scene. Same visual quality. A fraction of the memory.
For anyone who’s been watching VRAM capacities stagnate on mid-range cards while game requirements keep climbing, those numbers land hard. The RTX 5060, for example, ships with 8 GB of VRAM at around $350. That’s not a lot of breathing room for modern titles, and NTC could change the math entirely.
How NTC actually works, and why it’s different from what you know
Right now, most games compress textures using a method called BCn, block compression. The GPU breaks textures into chunks, compresses them, and decompresses them as needed during rendering. It’s been the standard for years, and it works, but it has a ceiling when it comes to how much it can squeeze out of the data.
NTC takes a completely different approach. Instead of storing textures in their traditional form, it encodes them into what’s called a latent representation, a much more compact format, and then uses a small neural network running directly on the GPU to reconstruct the texture data in real time, on demand, during gameplay. The result is that same Tuscan Villa scene going from 6.5 GB down to 970 MB, and here’s the part that really stands out: at that 970 MB budget, NTC actually preserved more detail than regular block compression. Not just equivalent quality, better.
The practical benefits extend beyond VRAM. Smaller texture data means smaller game installs, lighter patches, and less download bandwidth. In an era where some games are pushing 100 GB or more on disk, that’s not a small thing.
Alongside NTC, NVIDIA also presented Neural Materials, which applies the same logic to how material properties are stored and processed. Instead of carrying a large stack of separate texture data and running heavy mathematical models to calculate how light interacts with surfaces, Neural Materials compresses that behavior into a compact representation that a small network decodes at render time.
In testing, a material setup that previously required 19 channels was reduced to just 8, and NVIDIA reported render times improving between 1.4x and 7.7x at 1080p. That kind of headroom frees up GPU resources for everything else happening in a scene.
No games support it yet, but the industry is moving
Here’s where expectations need to stay grounded. NTC is not new, NVIDIA first showed it nearly three years ago, and despite an SDK being available since early 2026, not a single shipped game uses it yet. That’s likely exactly why NVIDIA dedicated a full GTC session to it again: the tools exist, but developer adoption hasn’t happened.
What needs to fall into place is support for DirectX Cooperative Vectors, Microsoft’s standardized API that allows pixel shaders to use the same hardware acceleration behind neural network inference. The good news is that this technology isn’t locked to NVIDIA hardware. Microsoft has standardized it across DirectX, Intel has already shown its own demo with neural texture compression, and AMD discussed the technology back in 2024 and is expected to support it as well. This is shaping up to be an industry-wide shift, not a vendor exclusive.
There’s also a rumor circulating from leaker Kepler_L2 suggesting Sony could use NTC on the PlayStation 6 to reduce game install sizes while keeping storage costs down with a 1 TB SSD. Nothing official, but the pieces fit.
The beta SDK is sitting on GitHub right now, ready for developers to pick up. The technology works, the results are real, and unlike DLSS 5, which has stirred debate around AI-generated frames and artistic intent, NTC operates invisibly inside the pipeline. It shrinks assets, speeds up shading, and frees up memory without touching the final image or overriding any creative decision a developer made. It’s AI doing the unglamorous work so everything else can look better.
The only question is how long it takes studios to actually use it.
Think NTC is the real answer to the VRAM problem, or are you holding your breath until it actually shows up in a game? Tell us in the comments!