Nvidia Announces AI-Powered Dynamic Super Resolution image scaling technology
Improving image quality without the performance hit.
Nvidia is working on a new AI based image quality and scaling technology. Its upcoming driver, scheduled for release on January 14th, includes what Nvidia calls DLDSR (Deep Learning Dynamic Super Resolution). It’s quite different from Nvidia's current and quite popular scaling solution, DLSS. DLDSR renders a game at higher and therefore more detailed resolution using the Tensor cores of an RTX GPU before scaling back down to the resolution of your monitor. This means you’ll get enhanced image quality with higher detail and smoother edges.
Nvidia released a screenshot showing three side by side images, one at native 1080p, one using DSR at 4K>1080p and one using DLDSR at 1620p>1080p. Looking at the image, the boxes with Petersons written on them look sharper in the DSR image and better again on the DLDSR image. And at 145 FPS for the native shot vs 143 for the DLDSR shot, the performance impact is negligible.
How to buy a graphics card: tips on buying a graphics card in the barren silicon landscape that is 2021
As this is a driver level tech, Nvidia say it ‘works in most games’ and any RTX GPU with Tensor cores will support DLDSR.
At the same time, Nvidia announced a partnership with Pascal Gilcher, a modder and the author of the ReShade Ray Tracing Filter to produce three Freestyle filters for its GeForce Experience tool. Nvidia claims these filters apply lighting and shadow enhancements that look as good as an actual remastered experience. That’s a bold claim. It will be interesting to see how the tech works across a wide variety of games.
Image upscaling and downscaling tech is developing rapidly along with the acronyms to describe them. DLSS continues to evolve, and there’s also AMD’s RSR or Radeon Super Resolution. Intel is also working on its own XeSS technology.
DLDSR and the new Freestyle will be included in Nvidia’s driver that will be available on the 14th of January.