NotebookCHECK - Notebook Forum

English => News => Topic started by: Redaktion on April 06, 2026, 14:18:59

Title: Impressive results: Nvidia presents a possible solution to the VRAM crisis
Post by: Redaktion on April 06, 2026, 14:18:59
At GTC 2026, Nvidia is presenting a possible answer to the VRAM crisis in modern games: with Neural Texture Compression, memory usage in a demo drops from 6.5 GB to 970 MB while maintaining nearly identical image quality. The technology is giving the community hope, but it is also fueling criticism of Nvidia's past VRAM strategy.

https://www.notebookcheck.net/Impressive-results-Nvidia-presents-a-possible-solution-to-the-VRAM-crisis.1267085.0.html
Title: Re: Impressive results: Nvidia presents a possible solution to the VRAM crisis
Post by: one band on April 07, 2026, 08:24:33
"memory usage in a demo drops from 6.5 GB to 970 MB while maintaining nearly identical image quality"

Great! Can we do the same with their pay? We're going to drop your pay a large amount but somehow, you're going to be able to nearly buy what you could before.
Title: Re: Impressive results: Nvidia presents a possible solution to the VRAM crisis
Post by: Trent on April 07, 2026, 10:51:20
Didn't read the article yet, but let me guess. AI?
Title: Re: Impressive results: Nvidia presents a possible solution to the VRAM crisis
Post by: It can be a win on April 07, 2026, 12:25:59
Since NVIDIA started comparing apples to oranges, like in Ada vs Blackwell and Blackwell looked twice as fast, but they were comparing FP8 vs FP4, and it was only visible in small print: I believe it when I see it or when independent reviewers confirm it, but even if it's a 10% VRAM reduction in real games's content, that would be something (of course, if everything remains the same, that is).

So this texture compression happens in real time and requires, I assume, some kind of special cores to do the compression, which would require additional energy. (again, as said, if everything remains the same and it doesn't lead to less energy efficiency e.g.).

But it can be a win as DLSS upscaling generally has shown.