NVIDIA Cuts VRAM Use 85% with Neural Texture Compression
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- NVIDIA demonstrated Neural Texture Compression (NTC) at its GTC 2026 presentation, showing a reduction in VRAM usage from 6.5 GB to 970 MB on the Tuscan Wheels scene while maintaining near‑original image quality.
- NTC preserved more detail than traditional block‑compression (BCN) at the same 970 MB memory budget, indicating higher visual fidelity for the same memory footprint.
- Neural Materials compressed a 19‑channel material setup to eight channels, delivering 1.4×–7.7× faster 1080p render times in the demo.
- NVIDIA positioned neural rendering as a broader pipeline‑level approach distinct from DLSS 5, which only applies AI to the final rendered frame.
- Demo highlighted practical benefits such as smaller game installs, reduced patch sizes, lower download bandwidth, and more room for detailed assets on GPUs.
- Jensen Huang responded to DLSS 5 criticism, emphasizing that the new neural rendering techniques aim to improve optimization rather than alter artistic direction.
Why it matters: Gamers and developers can run higher‑detail textures without exceeding GPU VRAM limits, while publishers can shrink game installs, patches and download bandwidth, cutting costs. The technique also shows AI can optimize memory pipelines, expanding the role of neural rendering beyond DLSS 5’s image upscaling.
