FSR explaination PLEASE

[majin]

when it comes to fsr im a literal donkey. i didnt unterstand it in spiderman games either. should i turn it on? yes? if yes balanced? no? i remember changing this stupid setting for 30 minutes straight to see how spiderman would run better and now my worst enemy is here again. can someone explain it to my donkeybrain please

[ducklegs]

If you have NVIDIA RTX card, turn on DLSS to quality or balanced. If you don't, turn FSR to quality or balanced

[Mazzle Dazzle]

I personally turn FSR onto performance on every game and run at 1080p. My graphics card can't handle anything more than 1080p even on my desktop alone.

[Elenoe]

FSR (AMD) and DLSS (NVIDIA) are upscaling/time interpolation algorithms. DLSS much better then FSR. Tho FSR is simpler and will run on anything. DLSS actually requires RTX card, and higher tier at that to fully utilize it.

Ideally you want to have them disabled, if your GPU can handle the raw power required for your target resolution@framerate (like 4K@60) you are happy camper.

But if it can't, you can turn them on. They both effectivelly start rendering the image in lower resolution then what you have set. And upscales it to your desired resolution thus getting you more FPS for a lower quality image.

The Quality vs Balanced vs Performance refers to the level of upscaling. Like (just an example as it depends on resolution too) DLSS Quality uses 66% of target reslution. While Performance uses 50% of target resolution.

--------

Worth noting the upscaling method is rather sophisticated. Without going into specifics DLSS for example uses the fact, that the scene your are watching doesn't really change 60x times per seconds. 90% of the image remains more or less the same, just shifted, observed from a slightly different angle. The objects you see are mostly the same. So it can "borrow" the missing parts from the previously rendered frames while the "AI" predicts "what's changed" without actually needing the GPU to give the fully rendered results.

So even if you play on half resolution (meaning 1/4th of pixels rendered) the image still looks very close to 4K. The better the less you are moving. The more erratic moves you will do the more "blurred" it would appear. But that's how eye works as well so normally it's hard to notice unless you focus on some effects like ray tracing shades, or edges on near passing objects if you move fast around them.

The newest (DLSS3 with Frame Generation) and the experimental gen (DLSS 3.5 with Ray Reconstruction), only running on 4xxx RTX cards, can do that to fill a complete frames with expensive effects. It only renders the easy part of the scene. But the hard part, like ray tracing, global illumination etc.. it only computes at low FPS and "predicts" the missing parts based on the scene dynamics. It can be seen in Cyberpunk 2.0. Everyone will swear the game runs 100FPS (and the most important parts do) but the true framerate can go as low as 30FPS without it starting to be uncomfortable or having as much artifacts as with previous versions.

FSR on the other hand is much less taxing but way worse. It does the same job tho. Upscaling images and (since FSR3) frame generation. It however only uses the images themselves and not the vital extra info about model changes from the partial rendering. Which results in worse image. But it will run on low end cards and it will still give you more FPS without the need to lower resolution.