Hey, would this not bea great ? We could render at HALF the resolution improving speed by 4x.
???
1 Like
DLSS is per game or per app, as it needs to be trained. That by definition already makes it unsuitable for a general 3D app like Blender. The training is also done by Nvidia on game to game basis, who also owns the technology and has a complete control over it.
DLSS was per game, DLSS 2.0 is not anymore. The model is trained for any content or game. Also like Optix, Nvidia can bring it to Blender too. There was a heavy help from Nvidia to implement Optix and Denoise into Blender.
6 Likes
https://github.com/dynamiquel/NVIDIA-Ansel-AI-Enhancer … its not the same as dlss i guess but its just as new … works good on aliased stuff like hair or alphas imo. needs rtx 20xx+ though , there are several other upscaling gans on github if you search , most have to be called from python commandline though
1 Like
I appreciate your enthusiasm, but you’ve bumped multiple topics recently that are several years old, and bumping is discouraged in forum rules. It’s annoying to check the new topics list only to find threads that are outdated and no longer relevant. You obviously don’t have to listen to me, I have no authority at all, I’m just asking you as a courtesy to be mindful of that date a topic was last responded to
Actually this is part of what is best for blender - and it’s not about when it was posted.
AMD FSR is a better bet though as it works for both
1 Like
That is something for your UPBGE devs to implement after they have implemented vulkan or dx12 support.
1 Like