If you're anything like me, you'll be bouncing between graphics settings more often than you might like.
www.pcgamer.com
Clouds aside, this article actually touches on an issue I don't think folks appreciate enough when they are bullying companies about supporting every upscaler:
A pretty handy feature, I think you'll agree, although it inadvertently exposes an uncomfortable truth—each upscaler is good at different things, and that makes choosing the right one a very subjective process.
I wouldn't consider myself an expert in upscalers, but I've looked at the various upscaling outputs in detail in some of these noisier cases quite a bit as you would imagine as part of my job. While they all have been getting somewhat better over time, they
all have various cases where they break down, sometimes in pretty bad ways. ex. TSR tends to have some issues with high contrast shadows/specular (as shown in one case in the DF hellblade video). DLSS has various issues with excessive ghosting and generally doesn't handle transparencies super well. FSR2 tends to leave a fair bit of noise in the image even with relatively little motion. Etc. Don't take these as gospel because this stuff changes all the time; the point is just that all of them have places they work well and places they fall apart a bit - none of them is magical.
Given the rather significant effect on the output pixels of these techniques, it seems a little crazy to keep treating this like it's a user settings problem. Putting a black box in the API (or SDKs like "streamline") often makes this even worse because now it's not only a black box of a few techniques that you have to try and QA, but a black box of "arbitrary future implementations on arbitrary future hardware" that could completely screw up the image and the game developer has no control or no way to QA that.
Games really need a way to be able to test and see the real pixels that are output and once QA signs off, those are the pixels that should show up for the user, until the end of time ideally. The test matrix with the current user demands to support every single upsampling technique is already getting to be a problem since there are so many cases that they need to be tested in. Issues like the clouds here will only get worse as it's just unreasonable to expect with so many moving parts controlled by third parties that there aren't going to be problems. Is this a game problem? Is this an API problem? Is it a DLSS problem? It's none and all of them at the same time, with no clear owner to maintain compatibility and "fixes".
Of course I realize the commercial reality of this direction, but I will not hesitate to continue to point out places where this giant black box that is the ultimate arbiter of the pixels you see at the end of the pipeline is a problematic model going forward.